WO2023155041A1 - 一种智能驾驶方法、装置及包括该装置的车辆 - Google Patents

一种智能驾驶方法、装置及包括该装置的车辆 Download PDF

Info

Publication number
WO2023155041A1
WO2023155041A1 PCT/CN2022/076331 CN2022076331W WO2023155041A1 WO 2023155041 A1 WO2023155041 A1 WO 2023155041A1 CN 2022076331 W CN2022076331 W CN 2022076331W WO 2023155041 A1 WO2023155041 A1 WO 2023155041A1
Authority
WO
WIPO (PCT)
Prior art keywords
vehicle
decision
self
processor cores
tasks
Prior art date
Application number
PCT/CN2022/076331
Other languages
English (en)
French (fr)
Inventor
卢远志
邱梅清
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to PCT/CN2022/076331 priority Critical patent/WO2023155041A1/zh
Publication of WO2023155041A1 publication Critical patent/WO2023155041A1/zh

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W60/00—Drive control systems specially adapted for autonomous road vehicles
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/16—Anti-collision systems

Definitions

  • the present application relates to the technical field of intelligent driving, in particular to a method and device based on intelligent driving and a vehicle including the device.
  • Intelligent driving can also be called automatic driving or assisted driving. It is an important direction for the development of intelligent vehicles. More and more rich driving functions, and gradually realize different levels of driving experience.
  • the Society of Automotive Engineers (SAE) provides a classification standard for driving automation, including driving levels L0 to L5, where L0 is no automation, the human driver has full authority to operate the car, and can be driven during driving.
  • System warning or assistance such as automatic emergency braking (autonomous emergency braking, AEB), blind spot detection (blind spot monitoring, BSM) or lane departure warning (lane departure warning, LDW), etc.
  • L1 level is driving support. The driving operation is completed by the human driver and the driving system.
  • the driving system can provide driving support for the steering wheel or acceleration and deceleration operations through the driving environment.
  • Other driving operations are performed by the human driver, such as adaptive cruise control (Adaptive cruise control (ACC) or lane keep assistance/support (LKA/LKS), etc.
  • L2 level is partially automated, providing driving support for steering wheel and acceleration and deceleration through the driving environment, and other driving Actions are performed by human drivers, such as car-following functions that combine adaptive cruise control (ACC) and lane keep assistance (LKA);
  • L3 is conditional automation, which can be done by the driving system
  • the driving operation but the human driver needs to respond to the request of the driving system at an appropriate time, that is, the human driver needs to be ready to take over the driving system;
  • L4 level is highly automated, and all driving operations can be completed by the driving system.
  • the human driver does not necessarily need to respond to the request of the driving system.
  • the road and environmental conditions permit (such as closed parks, highways, urban roads or fixed driving routes, etc.)
  • the human driver may not take over the driving;
  • L5 The level is fully automated, and the driving operation under various road and environmental conditions that human drivers can cope with can be completed by the driving system autonomously. It can be seen that at the level of L0 to L2, the driving system mainly provides support for the driver, and the driver still needs to do a good job of driving supervision, and steer, brake or accelerate as needed to ensure safety. From L3 to L5, the driving system can replace the driver to complete all driving operations. At the L3 level, the driver must be ready to take over the driving. At the L4 and L5 levels, the driving system can realize complete driving under some conditions and all conditions. Members can choose whether to take over.
  • an intelligent driving method which can enable the vehicle's automatic driving system to smoothly and reliably handle multiple dynamic target tracking road scenes (especially left-turn scenes), and improve the safety of the vehicle's automatic driving.
  • an automatic driving device is provided, and the automatic driving device can be used in vehicles with intelligent driving capabilities (such as the above-mentioned vehicles with L2-L5 automatic driving capabilities).
  • the automatic driving device includes: a processing device, the processing device includes a plurality of processor cores; at least a part of the processor cores in the plurality of processor cores are fixedly assigned to a set task set, and the fixed assignment refers to: the plurality of processing At least a part of processor cores in the processor cores are configured to only process a set task set, and the set task set is a non-empty set.
  • the automatic driving device further includes a perception system for obtaining the perception information of the own vehicle, and the perception system includes at least one of the following: laser radar, millimeter wave radar, camera, ultrasonic radar, on vehicles with intelligent driving capabilities , multiple sensors of the same type can be set, for example, two or three lidars can be set on the vehicle.
  • the set task set includes a sensing task, and the sensing task includes processing information acquired by the sensing system by the at least a part of the processor cores and acquiring environmental information and obstacle information around the vehicle.
  • the obstacle information includes at least one of the following: static obstacle information and dynamic obstacle information.
  • Static obstacles may include, for example, houses on the side of the road, vehicles parked on the side of the road, lampposts, trash cans, etc.
  • dynamic obstacles may include, for example, moving vehicles, pedestrians, and the like.
  • the automatic driving device further includes a planning system; the set task set includes a planning task, and the planning task includes calculating the obstacle based on the movement information of the vehicle by the at least a part of the processor cores. Determine the planned driving strategy of the vehicle based on the object information.
  • the sensing task includes at least one of the following: a lane line recognition (LD) task for identifying lane lines around the vehicle, a vehicle detection (VD) task for identifying other vehicles around the vehicle, and combining different sensor information Carry out fusion perception fusion (RL), perceive and track the key targets around the vehicle, the key targets include vulnerable road users (VRU); the planning task includes at least one of the following: global path planning of the vehicle (Router ), Finite State Machine (FSM), Motion Control (MOP).
  • LD lane line recognition
  • VD vehicle detection
  • RL Carry out fusion perception fusion
  • VRU vulnerable road users
  • the planning task includes at least one of the following: global path planning of the vehicle (Router ), Finite State Machine (FSM), Motion Control (MOP).
  • At least one processor core in the at least a portion of the processor cores is fixedly allocated to one of the sensing tasks. And/or fixedly assigning at least one processor core in the at least a part of processor cores to one of the planned tasks.
  • the automatic driving task it mainly includes perception tasks and planning tasks.
  • the planning system plans the formal strategy of the vehicle based on the information acquired by the perception system, by assigning at least a part of the multiple processor cores to the perception tasks and/or
  • the planning task can ensure the smooth completion of the planning task in the process of vehicle automatic driving.
  • At least one processor core in at least a portion of the processor cores is fixedly assigned to perceive and track key targets around the vehicle
  • the key targets include Vulnerable Road Users (VRUs)
  • VRUs Vulnerable Road Users
  • safety is the first priority task.
  • the automatic driving device obtains the decision-making scheme set of the self-vehicle according to the perception information; and obtains the first decision-making scheme for the first interaction stage between the self-vehicle and the game target from the decision-making scheme set;
  • the process of controlling the self-vehicle to drive by the first decision-making scheme it is determined that the self-vehicle and the game target meet the conditions for entering the second interaction stage;
  • the game object performs a second decision-making scheme in the second interaction stage; the self-vehicle is controlled to drive based on the second decision-making scheme.
  • the vehicle can make an appropriate automatic driving strategy in real time according to the surrounding conditions.
  • the first interaction stage is a stage in which the interaction relationship between the ego vehicle and the game object is not clear;
  • the second interaction stage is a stage in which the interaction relationship between the ego vehicle and the game object is clear.
  • the automatic driving device is configured to: determine at least one target according to the perception information; acquire the future trajectory of the at least one target; acquire the future trajectory of the self-vehicle; The target whose future trajectory intersects with the future trajectory of the self-vehicle is determined as the game target, and during the first interaction phase between the self-vehicle and the game target, the game target Status is continuously tracked.
  • the vehicle can continue to pay attention to the game goals (that is, goals that may conflict with the vehicle in the future), thereby reducing or even discontinuing the allocation of processing
  • the computing power of the core is on the non-game goal (that is, the goal that will not conflict with the vehicle in the future), so that computing power can be saved.
  • it includes an intelligent driving vehicle, which includes the automatic driving device in the above-mentioned various embodiments.
  • Figure 1 shows a schematic diagram of the composition of an intelligent driving vehicle
  • Fig. 2 shows a schematic framework diagram of an automatic driving system according to an embodiment of the present application
  • Figure 3-1 shows a schematic diagram of a common scene in automatic driving
  • Figure 3-2 shows a schematic diagram of a common scene in automatic driving
  • Figure 4 shows the number of targets tracked by a self-driving vehicle over a period of time in an actual scene
  • Fig. 5 shows a schematic diagram of a common scene in automatic driving
  • Fig. 6 shows a schematic diagram of a common scene in automatic driving
  • Fig. 7 shows a schematic diagram of a common scene in automatic driving
  • Fig. 8 shows a schematic diagram of a common scene in automatic driving
  • Fig. 9 shows a schematic diagram of a common scene in automatic driving
  • Fig. 10 shows a schematic diagram of a processor core of an automatic driving system
  • Figure 11 shows a schematic diagram of a common scene in automatic driving
  • Fig. 12 shows a schematic diagram of determining the game target of the self-vehicle according to the embodiment of the present application
  • Fig. 13 shows a schematic diagram of the principle of determining the decision-making solution set of the own vehicle according to the embodiment of the present application
  • Fig. 14 shows a schematic diagram of labeling the own vehicle decision-making solution set according to an embodiment of the present application
  • Fig. 15 shows a schematic diagram of evaluation dimensions for evaluating the benefits of the decision-making scheme of the self-vehicle according to an embodiment of the present application
  • FIG. 16 shows a schematic diagram of an offset interval according to an embodiment of the present application.
  • Fig. 17 shows a schematic diagram of processing the ego vehicle action sequence according to an embodiment of the present application
  • Fig. 18 shows a schematic flowchart of the interactive game between the ego vehicle and the game object according to the embodiment of the present application.
  • autonomous driving or “autonomous driving vehicle” referred to in the subsequent part of this application refers to vehicles with L2 and above capabilities classified according to SAE.
  • FIG. 1 is a schematic functional block diagram of a vehicle 100 .
  • Vehicle 100 may be configured in a fully or partially autonomous driving mode.
  • the vehicle 100 can obtain its surrounding environment information through the perception system 120, and obtain an automatic driving strategy based on the analysis of the surrounding environment information to realize fully automatic driving, or present the analysis results to the user to realize partially automatic driving.
  • Vehicle 100 may include various subsystems such as infotainment system 110 , perception system 120 , decision control system 130 , drive system 140 , and computing platform 150 .
  • vehicle 100 may include more or fewer subsystems, and each subsystem may include multiple components.
  • each subsystem and component of the vehicle 100 may be interconnected in a wired or wireless manner.
  • the infotainment system 110 may include a communication system 111 , an entertainment system 112 and a navigation system 113 .
  • Communication system 111 may include a wireless communication system that may wirelessly communicate with one or more devices, either directly or via a communication network.
  • wireless communication system 146 may use 3G cellular communications, such as CDMA, EVDO, GSM/GPRS, or 4G cellular communications, such as LTE. Or 5G cellular communications.
  • the wireless communication system may communicate with a wireless local area network (wireless local area network, WLAN) by using WiFi.
  • the wireless communication system 146 may communicate directly with the device using an infrared link, Bluetooth, or ZigBee.
  • Other wireless protocols, such as various vehicle communication systems, for example, a wireless communication system may include one or more dedicated short range communications (DSRC) devices, which may include communication between vehicles and/or roadside stations Public and/or Private Data Communications.
  • DSRC dedicated short range communications
  • the entertainment system 112 can include a central control screen, a microphone and a sound system. Users can listen to the radio and play music in the car based on the entertainment system; Touch type, users can operate by touching the screen. In some cases, the user's voice signal can be acquired through the microphone, and the user can control the vehicle 100 based on the analysis of the user's voice signal, such as adjusting the temperature inside the vehicle. In other cases, music may be played to the user via a speaker.
  • the navigation system 113 may include a map service provided by a map provider, so as to provide navigation for the driving route of the vehicle 100 , and the navigation system 113 may cooperate with the global positioning system 121 and the inertial measurement unit 122 of the vehicle.
  • the map service provided by the map provider can be a two-dimensional map or a high-definition map.
  • the perception system 120 may include several kinds of sensors that sense information about the environment around the vehicle 100 .
  • the perception system 120 may include a global positioning system 121 (the global positioning system may be a GPS system, or a Beidou system or other positioning systems), an inertial measurement unit (inertial measurement unit, IMU) 122, a laser radar 123, a millimeter wave radar 124 , ultrasonic radar 125 and camera device 126 .
  • the perception system 120 may also include sensors of the interior systems of the monitored vehicle 100 (eg, interior air quality monitors, fuel gauges, oil temperature gauges, etc.). Sensor data from one or more of these sensors can be used to detect objects and their corresponding properties (position, shape, orientation, velocity, etc.). Such detection and identification is a critical function for safe operation of the vehicle 100 .
  • the global positioning system 121 may be used to estimate the geographic location of the vehicle 100 .
  • the inertial measurement unit 122 is used to sense the position and orientation changes of the vehicle 100 based on inertial acceleration.
  • inertial measurement unit 122 may be a combination accelerometer and gyroscope.
  • the lidar 123 may utilize laser light to sense objects in the environment in which the vehicle 100 is located.
  • lidar 123 may include one or more laser sources, a laser scanner, and one or more detectors, among other system components.
  • the millimeter wave radar 124 may utilize radio signals to sense objects within the surrounding environment of the vehicle 100 .
  • radar 126 may be used to sense the velocity and/or heading of objects.
  • the ultrasonic radar 125 may sense objects around the vehicle 100 using ultrasonic signals.
  • the camera device 126 can be used to capture image information of the surrounding environment of the vehicle 100 .
  • the camera device 126 may include a monocular camera, a binocular camera, a structured light camera, a panoramic camera, etc., and the image information acquired by the camera device 126 may include still images or video stream information.
  • the decision-making control system 130 includes a computing system 131 for analyzing and making decisions based on the information acquired by the perception system 120.
  • the decision-making control system 130 also includes a vehicle controller 132 for controlling the power system of the vehicle 100, and for controlling the steering of the vehicle 100.
  • Computing system 131 is operable to process and analyze various information acquired by perception system 120 in order to identify objects, objects, and/or features in the environment surrounding vehicle 100 .
  • the objects may include pedestrians or animals, and the objects and/or features may include traffic signals, road boundaries, and obstacles.
  • the computing system 131 may use techniques such as object recognition algorithms, structure from motion (SFM) algorithms, and video tracking. In some embodiments, computing system 131 may be used to map the environment, track objects, estimate the velocity of objects, and the like.
  • the computing system 131 can analyze various information obtained and obtain a control strategy for the vehicle.
  • the vehicle controller 132 can be used for coordinated control of the power battery and the engine 141 of the vehicle, so as to improve the power performance of the vehicle 100 .
  • the steering system 133 is operable to adjust the heading of the vehicle 100 .
  • it could be a steering wheel system.
  • the throttle 134 is used to control the operating speed of the engine 141 and thus the speed of the vehicle 100 .
  • the braking system 135 is used to control deceleration of the vehicle 100 .
  • Braking system 135 may use friction to slow wheels 144 .
  • braking system 135 may convert kinetic energy of wheels 144 into electrical current.
  • the braking system 135 may also take other forms to slow the wheels 144 to control the speed of the vehicle 100 .
  • Drive system 140 may include components that provide powered motion to vehicle 100 .
  • drive system 140 may include engine 141 , energy source 142 , transmission 143 and wheels 144 .
  • the engine 141 may be an internal combustion engine, an electric motor, an air compression engine or other types of engine combinations, such as a hybrid engine composed of a gasoline engine and an electric motor, or a hybrid engine composed of an internal combustion engine and an air compression engine.
  • Engine 141 converts energy source 142 into mechanical energy.
  • Examples of energy source 142 include gasoline, diesel, other petroleum-based fuels, propane, other compressed gas-based fuels, ethanol, solar panels, batteries, and other sources of electrical power.
  • the energy source 142 may also provide energy to other systems of the vehicle 100 .
  • Transmission 143 may transmit mechanical power from engine 141 to wheels 144 .
  • Transmission 143 may include a gearbox, a differential, and a drive shaft.
  • the transmission device 143 may also include other devices, such as clutches.
  • drive shafts may include one or more axles that may be coupled to one or more wheels 121 .
  • Computing platform 150 may include at least one processor 151 that may execute instructions 153 stored in a non-transitory computer-readable medium such as memory 152 .
  • computing platform 150 may also be a plurality of computing devices that control individual components or subsystems of vehicle 100 in a distributed manner.
  • Processor 151 may be any conventional processor, such as a commercially available CPU.
  • the processor 151 may also include, for example, an image processor (Graphic Process Unit: GPU), a field programmable gate array (Field Programmable Gate Array: FPGA), a system on a chip (Sysem on Chip: SOC), an application-specific integrated chip ( Application Specific Integrated Circuit: ASIC) or their combination.
  • FIG. 1 functionally illustrates the processor, memory, and other elements of computer 110 in the same block, those of ordinary skill in the art will understand that the processor, computer, or memory may actually include Multiple processors, computers, or memories stored within the same physical enclosure.
  • the memory may be a hard drive or other storage medium located in a different housing than the computer 110 .
  • references to a processor or computer are to be understood to include references to collections of processors or computers or memories that may or may not operate in parallel.
  • some components such as the steering and deceleration components, may each have their own processor that only performs calculations related to component-specific functions .
  • the processor may be located remotely from the vehicle and be in wireless communication with the vehicle. In other aspects, some of the processes described herein are executed on a processor disposed within the vehicle while others are executed by a remote processor, including taking the necessary steps to perform a single maneuver.
  • memory 152 may contain instructions 153 (eg, program logic) executable by processor 151 to perform various functions of vehicle 100 .
  • Memory 152 may also contain additional instructions, including sending data to, receiving data from, interacting with, and/or controlling one or more of infotainment system 110 , perception system 120 , decision control system 130 , drive system 140 instructions.
  • memory 152 may also store data such as road maps, route information, the vehicle's position, direction, speed, and other such vehicle data, among other information. Such information may be used by vehicle 100 and computing platform 150 during operation of vehicle 100 in autonomous, semi-autonomous, and/or manual modes.
  • Computing platform 150 may control functions of vehicle 100 based on input received from various subsystems (eg, drive system 140 , perception system 120 , and decision-making control system 130 ). For example, computing platform 150 may utilize input from decision control system 130 in order to control steering system 133 to avoid obstacles detected by perception system 120 . In some embodiments, computing platform 150 is operable to provide control over many aspects of vehicle 100 and its subsystems.
  • various subsystems eg, drive system 140 , perception system 120 , and decision-making control system 130 .
  • computing platform 150 may utilize input from decision control system 130 in order to control steering system 133 to avoid obstacles detected by perception system 120 .
  • computing platform 150 is operable to provide control over many aspects of vehicle 100 and its subsystems.
  • one or more of these components described above may be installed separately from or associated with the vehicle 100 .
  • memory 152 may exist partially or completely separate from vehicle 100 .
  • the components described above may be communicatively coupled together in a wired and/or wireless manner.
  • FIG. 1 should not be construed as limiting the embodiment of the present application.
  • An autonomous vehicle traveling on a road can identify objects within its surroundings to determine adjustments to the current speed.
  • the objects may be other vehicles, traffic control devices, or other types of objects.
  • each identified object may be considered independently and based on the object's respective characteristics, such as its current speed, acceleration, distance to the vehicle, etc., may be used to determine the speed at which the autonomous vehicle is to adjust.
  • the vehicle 100 or a sensing and computing device (e.g., computing system 131, computing platform 150) associated with the vehicle 100 may be based on the identified characteristics of the object and the state of the surrounding environment (e.g., traffic, rain, traffic on the road) ice, etc.) to predict the behavior of the identified objects.
  • each identified object is dependent on the behavior of the other, so all identified objects can also be considered together to predict the behavior of a single identified object.
  • the vehicle 100 is able to adjust its speed based on the predicted behavior of the identified object.
  • the self-driving car is able to determine what steady state the vehicle will need to adjust to (eg, accelerate, decelerate, or stop) based on the predicted behavior of the object.
  • other factors may also be considered to determine the speed of the vehicle 100 , such as the lateral position of the vehicle 100 in the traveling road, the curvature of the road, the proximity of static and dynamic objects, and the like.
  • the computing device may also provide instructions to modify the steering angle of the vehicle 100 such that the self-driving car follows a given trajectory and/or maintains contact with objects in the vicinity of the self-driving car (e.g., , the safe lateral and longitudinal distances of cars in adjacent lanes on the road.
  • objects in the vicinity of the self-driving car e.g., , the safe lateral and longitudinal distances of cars in adjacent lanes on the road.
  • the above-mentioned vehicle 100 may be a car, truck, motorcycle, bus, boat, airplane, helicopter, lawn mower, recreational vehicle, playground vehicle, construction equipment, tram, golf cart, train, etc., the embodiment of the present application There is no particular limitation.
  • Fig. 2 shows a schematic framework diagram of an automatic driving system in some embodiments.
  • the automatic driving system mainly includes a perception module, a planning module and a control module.
  • the automatic driving system can also be called an automatic driving system (automated driving system, ADS) or a driving assistance system, for example, an advanced driving assistance system (advanced driving assistant system, ADAS).
  • ADS automatic driving system
  • ADAS advanced driving assistance system
  • the automatic driving system uses the sensors on the vehicle to obtain the information of the vehicle itself and the information around the vehicle, and analyzes and processes the acquired information to realize, for example, obstacle perception, target recognition, vehicle positioning, path planning, driver monitoring/reminder, etc. function, thereby improving the safety, automation and comfort of vehicle driving.
  • the perception module has the ability of environment perception, such as identifying obstacles around the vehicle, detecting road markings, identifying signal lights, and detecting the behavior of pedestrians/vehicles around the vehicle, etc.
  • environment perception such as identifying obstacles around the vehicle, detecting road markings, identifying signal lights, and detecting the behavior of pedestrians/vehicles around the vehicle, etc.
  • sensors are installed on the vehicle, such as lidar, millimeter-wave radar, and cameras.
  • the perception module can also have the ability to perceive the vehicle itself, and there are more types of sensors for the perception of the vehicle's own state, including for measuring speed, temperature, pressure, flow, position, gas concentration, brightness, dry humidity, distance, etc. function of the sensor.
  • the perception module can obtain the speed of the vehicle through the speed sensor, and obtain the position of the accelerator pedal or the brake pedal through the position sensor.
  • the perception module can determine the position of the vehicle relative to the environment through a localization system.
  • the positioning system is, for example, a global positioning system (global positioning system, GPS) and/or an inertial navigation system (inertial navigation system).
  • the perception module obtains the information of the vehicle itself and the information around the vehicle through sensors, and can process the acquired information and provide it to the planning module. This processing includes, for example, information fusion of multiple sensors to improve the accuracy of the perception information .
  • the perception module may also be referred to as a perception fusion module.
  • a planning module that makes driving decisions/plans based on the information provided by the perception module. For example, according to the current environment information of the vehicle, a decision on the next driving behavior of the vehicle is made, such as acceleration/deceleration, lane change, steering, braking, or warning.
  • the automatic driving system can also realize information interaction with the on-board personnel (including the driver or passengers) through human-computer interaction, obtain the needs of the on-board personnel, and give feedback to the on-board personnel on the current status of the vehicle or alarms.
  • the planning module can also perform path decision/planning, for example, to complete the selection of the optimal path based on the user's needs, and the user's needs include, for example, one or more of the starting point, the end point, the route, and the route preference.
  • the control module makes corresponding controls based on the driving decision of the planning module, such as sending control instructions to the corresponding actuator or the controller of the actuator to control the actuator to perform corresponding actions, such as acceleration/deceleration, lane change, steering, braking , or warnings, etc.
  • the input of the automatic driving system not only includes the information input of its own sensors, but also can obtain information input from others to improve the performance of automatic driving.
  • the automatic driving system can obtain map information from the server of the map supplier to assist driving decision-making; another example, the automatic driving system can obtain driving environment information from the cloud, such as weather information, traffic status information (such as traffic flow, average traffic flow, etc.) Speed and other information) to assist driving decision-making; another example, the automatic driving system can obtain driving information and/or perception information of other cars from other smart terminals (such as other smart cars, pedestrians’ portable terminals, etc.) to assist driving decision-making.
  • the autonomous driving system can obtain information from other sources through wireless communication technology.
  • the wireless communication technology is, for example, cellular network-based communication or dedicated short range communications (DSRC), and the cellular network-based communication is, for example, long term evolution (long term evolution, LTE) communication technology or the fifth generation (5th generation) , 5G) communication technology.
  • communication based on cellular network includes vehicle-to-everything (V2X) communication
  • V2X communication includes vehicle-to-vehicle (V2V) communication, vehicle-to-roadside infrastructure ( Vehicle to infrastructure (V2I) communication, vehicle to pedestrian (V2P) communication or vehicle to network (V2N) communication, etc.
  • V2X communication includes vehicle-to-vehicle (V2V) communication, vehicle-to-roadside infrastructure ( Vehicle to infrastructure (V2I) communication, vehicle to pedestrian (V2P) communication or vehicle to network (V2N) communication, etc.
  • V2X communication includes vehicle-to-vehicle (V2V) communication, vehicle-to-roadside infrastructure ( Vehicle to infrastructure (V2I) communication, vehicle
  • an automatic driving system requires a large computing power, which is often realized through a dedicated hardware platform, such as a multi-domain controller (Multi Domain Controller, MDC), that is, all or part of the above perception module, planning module and control module
  • MDC Multi Domain Controller
  • the function is realized through a dedicated hardware platform, such as MDC.
  • the software in the automatic driving system can be upgraded, such as the algorithm of the perception module for information processing, or the decision algorithm of the planning module, etc. can be upgraded through software updates, and the software updates can be upgraded through the air (over the air, OTA) Technical realization.
  • smart driving vehicle generally refers to a vehicle with an SAE rating of L2 or above.
  • the "self-vehicle” is generally used to refer to the main vehicle including the automatic driving system. It should be understood that: the own-vehicle is a convenient descriptive designation designated by the vehicle where the driver is.
  • Figure 3-1 which exemplarily shows the left-turn scene of the self-vehicle at the intersection.
  • Four obstacles are shown, distributed as: A car, B car, pedestrian P1 and pedestrian P2.
  • the self-driving system of the vehicle will analyze these four obstacles and make their predicted trajectories respectively, as shown in Figure 1: the predicted trajectory of car A is going straight through the intersection from south to north, and the predicted trajectory of car B is The trajectory is going straight through the intersection from west to east, the predicted trajectory of pedestrian P1 is to pass the zebra crossing from north to south, the predicted trajectory of pedestrian P2 is to pass the zebra crossing from east to west, and the "predicted trajectory" of the ego vehicle turning left is shown in the figure .
  • Figure 3-2 shows the prediction of pedestrian P1 and the trajectory of pedestrian P1 by the automatic driving system of the own vehicle; it should be understood What is interesting is that although only one pedestrian P1 is shown in Figure 3-2, as mentioned above, in actual traffic scenarios, there may be a large number (such as dozens or even dozens of pedestrians crossing the zebra crossing); It can be seen that there are two possible conflict positions C1 and C2 between the predicted U-turn trajectory of the ego vehicle and the predicted trajectory of the pedestrian P1.
  • Figure 3-2 also shows the non-driving area A of the own vehicle.
  • the non-driving area A of the own vehicle is determined by the size of the vehicle itself and the turning radius of the vehicle, and cannot be entered when the own vehicle performs a U-turn.
  • Figures 3-1 and 3-2 are only some exemplary scenarios. In actual urban traffic scenarios, the number of obstacles is generally a dozen to dozens. Sometimes, it can even reach hundreds.
  • Figure 4 shows the number of targets (obstacles) tracked by an automatic driving vehicle in a certain period of time in the actual scene. It can be seen that the number of targets at the peak can almost reach one hundred, which will affect the automatic driving system of the vehicle. The demand for computing power produces a huge peak impact and leads to the above-mentioned various problems.
  • Figure 5 shows the scene of the vehicle driving out of the roundabout at the roundabout.
  • the own car plans to drive out of the roundabout at the next roundabout intersection according to the predicted trajectory.
  • car A in front of the left of the own car and C behind the right rear of the own car car, at another intersection there is car B that is about to enter the roundabout.
  • the predicted trajectory of the own vehicle may conflict with the predicted trajectory of the B vehicle and the predicted trajectory of the C vehicle; possibility of conflict.
  • Figure 5 shows the self-vehicle and the surrounding vehicles A, B, and C for example only.
  • the number of vehicles around the self-vehicle may be dozens or even tens of vehicles , and because in the roundabout scene, the exit/entrance of each vehicle driving out of the roundabout is inconsistent, which is more likely to lead to conflicts, therefore, the roundabout scene will also put forward real-time high computing power requirements for the self-driving system of the vehicle , this kind of high computing power demand often causes the self-driving system of the self-driving car to fail to operate normally and cause freezes, resulting in the self-driving car's self-driving system not being able to deal with the current situation in real time and in a timely manner, and unable to plan real-time and effective Autopilot strategy.
  • the ego vehicle In the T-junction scene as shown in Figure 7, the ego vehicle is expected to turn left at the intersection area, and the predicted trajectories of car A, car B, and pedestrians P1 and P2 are shown in Figure 7 respectively. It can be seen from Figure 7 that the ego vehicle The predicted left-turn trajectory of , may conflict with the above-mentioned car A, car B and pedestrians P1 and P2. Those skilled in the art should understand that Fig. 7 shows the self-vehicle and the surrounding A, B vehicles and pedestrians P1, P2 for the sake of example only. Ten vehicles, and the number of pedestrians may be dozens or even dozens.
  • the self-car is about to merge into the high-speed lane along the ramp, there is car A with the same intention on the ramp in front of the self-car, and there are cars B, C, D and E on the high-speed lane; the self-car's automatic
  • the tasks that the driving system needs to complete at the current moment include: judging the area where the ramp is about to end (that is, the road termination area in Figure 8), judging whether the predicted trajectory of the own car conflicts with car A, and judging the predicted trajectory of the own car and the high-speed lane Whether there is any conflict between cars B, C, D and E on the road, in the scene shown in Figure 8, cars B and C continue to drive along the current lane, and car D is slower because of the forward speed of car E , so it is planned to switch lanes, that is, merge into the lanes where cars B and C are located.
  • this scenario will impose real-time high computing power requirements on the self-driving system of the vehicle.
  • This high computing power This often causes the self-driving system of the self-vehicle to fail to operate normally, causing the self-driving system to fail to deal with the current situation in real time and in a timely manner, and to plan a real-time and effective self-driving strategy.
  • the ego vehicle can determine the lane termination area through the perception system
  • the perception system can include a camera and a laser radar
  • the perception system is based on the image information acquired by the camera device and the point cloud information acquired by the laser radar
  • Carry out sensor information fusion to judge the end of the lane, and plan the trajectory of the own vehicle in advance based on the distance and orientation between the end of the lane and the vehicle.
  • the acquisition of the steering intention of the ego vehicle for the above-mentioned car D can be obtained based on the turn signal information of the car D, or based on V2X (Vehicle to Everything)/V2V (Vehicle to vehicle) technology, or by
  • the self-driving system of the self-car determines the driving information of the D car for a period of time before the current moment (for example: if there is an overtaking behavior in the historical motion system within the perception range of the self-car's perception system and it has been kept high speed, and the vehicle in front of car D maintains a relatively low speed at the current moment, the self-driving system can think that car D has a strong overtaking intention, and predict the behavior of car D based on this strong overtaking intention. Future forecast trajectory), or based on a combination of the above-mentioned ways to determine.
  • FIG. 9 it shows a schematic diagram of a scene where the ego vehicle is about to get off the ramp from the expressway.
  • the own car plans to change lanes from the second lane on the right to the rightmost lane and get off the ramp; in the rightmost lane, there are cars A and C driving along the current lane, and in front of the lane where the own car is located, there is car B planning to Change right into the rightmost lane and exit the ramp. Since car B is in front of car A and car C, the off-ramp behavior of car B will affect car A and car C, and further affect the future trajectory of the own car. It should be understood that although Fig.
  • the main innovations of this application include: (1) fixedly match the tasks of the automatic driving system to the processor core of the automatic driving system according to different types (2) monitor the surroundings of the vehicle The obstacles adopt the interactive game processing strategy
  • the vehicle's automatic driving system When the vehicle's automatic driving system is running, it generally includes different task types: for example: operating system tasks, automatic driving tasks, data storage tasks, data uploading tasks to the cloud, resource scheduling tasks, etc.; see Figure 2 and related Note that autonomous driving tasks can be divided into perception tasks, decision/planning tasks and control tasks.
  • each current automatic driving system generally includes multiple processors, and each processor may include multiple processor cores.
  • Fig. 10 shows a schematic diagram of a processor core of an automatic driving system.
  • the automatic driving system includes dual SOC chips A and B, and each chip A and B includes 16 cores. Therefore, the automatic driving system includes a total of 16 cores. 32 processor cores.
  • the tasks of the automatic driving system are fixedly allocated/matched to the above-mentioned 32 processors according to the types, and can be allocated according to the following priorities, (1) operating system tasks, (2) automatic driving tasks, (3) Data transfer (storage) tasks and data cloud tasks, (4) Resource scheduling tasks and resource allocation tasks, etc. More specifically: for example, No. 0-1 processor cores of SOC A and B can be assigned to operating system tasks; No. 2-13 processor cores of SOC A and B can be assigned to automatic driving tasks; Processor core No. 14 is assigned to data storage and data cloud tasks; processor core No. 15 of SOC A and B is assigned to resource scheduling tasks and resource configuration tasks.
  • the automatic driving system may include three processors, each of which includes 16 processor cores; another example may be a fixed allocation of three processor cores for resource scheduling tasks and resource configuration tasks.
  • the above-mentioned "tasks of the automatic driving system are fixedly assigned to the processor cores of the automatic driving system” means: in terms of the above-mentioned automatic driving tasks, a total of 24 processor cores (No. 2-13 of SOC A and B) are allocated processor cores), when the vehicle is driving, the automatic driving tasks (perception, regulation and control) of its automatic driving system are all processed by these 24 processor cores; the automatic driving tasks will not be handled by other processor cores. In addition, other non-autonomous driving tasks (such as operating system, resource scheduling, etc.) will not be processed by these 24 processor cores.
  • the automatic driving task is assigned the largest number of processor cores (24); through this setting, it can be guaranteed that in some scenarios, when the number of targets (obstacles) is large At this time, the high computing power demand for the automatic driving system can be met.
  • the requirements for the tasks of the automatic driving system mainly come from the real-time dynamic tracking of the target obstacles, and the planning based on the real-time dynamic tracking of the target obstacles.
  • the proposed automatic driving strategy that is, the main task at this time is in the "perception" and "planning" parts of the automatic driving task.
  • processors are fixedly assigned to non-autonomous driving tasks, such as operating system and data storage tasks, etc.
  • non-autonomous driving tasks such as operating system and data storage tasks, etc.
  • the number of processor cores fixedly assigned to these tasks is small
  • the fixed number of processor cores assigned to autonomous driving tasks but still sufficient to handle the above tasks.
  • these non-automatic driving tasks are also highly demanded. For example, when there are many targets (obstacles) around the vehicle, the The amount of data to be stored also increases.
  • the computing power requirements of these non-automatic driving tasks will not extend beyond the fixedly allocated processor cores, thereby affecting the computing power requirements of the automatic driving tasks.
  • the automatic driving tasks also include, for example, GNSS (Global Navigation Satellite System) positioning tasks, map (or high-precision map) service tasks, and the like.
  • GNSS Global Navigation Satellite System
  • map or high-precision map
  • autonomous driving tasks can be prioritized and real-time requirements Set the priority of high perception and planning tasks to high, and set the priority of positioning and map services to medium.
  • the tasks with higher priority are given priority.
  • the fixedly assigned processor cores will prioritize “emergency” tasks (such as perception tasks and planning tasks) to ensure the smooth progress of autonomous driving tasks and vehicle safe driving.
  • the perception task and the planning task can be further divided based on the functions they specifically implement.
  • perception tasks it can include, for example, LD (Lane Detection: lane line detection), PD (Pedestrian Detection: pedestrian detection), VD (Vehicle Detection: vehicle detection), RL (Radar and LiDar perception fusion), etc.; to plan tasks
  • plan tasks it can include Router (global path planning), FSM (Finite Status Machine: finite state machine), MOP (Motion Planning: motion control), etc.
  • processor cores may be further fixedly allocated; for example, two of the above 24 processor cores may be fixedly allocated to the LD function, and one of the above 24 processor cores may be fixedly allocated to the MOP function. It should be understood that if the number of the above-mentioned specifically implemented functions is too large to satisfy the requirement that each specifically implemented function be fixedly allocated to at least one processor core. Then, the functions implemented above can be sorted according to the priority. For high-priority functions, the corresponding processor cores are fixedly allocated, and for low-priority tasks, the processor cores may not be fixedly allocated.
  • the "high priority" or “low priority” tasks in the above-mentioned automatic driving tasks can be pre-set, can also be manually set according to needs, and can also be based on the vehicle It is automatically adjusted according to the environment in which it is located. It should also be understood that the number of processor cores in the above-mentioned 24 processor cores fixedly assigned to the automatic driving task is only an example, and may be other numbers in practice, such as 36, 48 or other suitable Natural number.
  • processor cores For example, still taking the fixed assignment of 24 processor cores to the automatic driving task as an example, it can be pre-set before the automatic driving system is installed, and the two processor cores are fixedly assigned to the VD (vehicle detection) instead of the PD task. Allocate processor cores.
  • a user setting interface can be set in the autopilot system, and the user can set one or two fixed-allocation processor core tasks. For example, if the user thinks that the RL task is more important, one processor core can be fixedly allocated to the RL. Task.
  • PD peer detection
  • the PD task is a basic and important task. Therefore, in high-speed scenarios, processor cores may not be fixedly allocated to PD tasks; in urban areas, processor cores may be fixedly allocated to PD tasks.
  • presetting before the automatic driving system is loaded, and retain the function of automatic adjustment according to the scene where the vehicle is located.
  • FIG. 11 it exemplarily shows the situation in a crossroad scene.
  • the own vehicle intends to turn left and park in the left-turn waiting area.
  • the perception system of the own vehicle needs to perceive the target (obstacle )include:
  • Vehicles around the self-vehicle (car A, car B, car C, car D, car E, car F, car G; the above-mentioned cars A-G are only examples, and the number of vehicles around the self-car may be dozens or even dozens of vehicles); when there are larger vehicles around the own vehicle (not shown in Figure 8), the larger vehicles may also cause blind spots for the own vehicle, such as blocking the camera device of the own vehicle to obtain traffic lights signal of;
  • Pedestrian P1 (pedestrian P1 in Figure 8 is only an example, and the number of pedestrians around the vehicle may be dozens or even dozens of people in actual scenarios);
  • stationary obstacles around the road e.g. trees
  • Roadside equipment RSU (Road-Side Unit) around the road;
  • Stationary obstacles on the road such as gravel, potholes, etc.
  • the planning task of the self-vehicle will require automatic driving strategy planning based on the information of the perception system.
  • Factors to be considered in strategy planning include:
  • the ego vehicle needs to avoid stationary obstacles on the road and roadblocks/construction signs;
  • the planning system of the automatic driving system also needs to process a large amount of information to make a corresponding automatic driving strategy.
  • the planning system of the self-vehicle needs to avoid stationary obstacles and roadblocks/construction signs when planning the path; it needs to avoid collisions with surrounding pedestrians and two-wheeled vehicles; it needs to avoid conflicts with the trajectories of other vehicles A-G around it, etc. wait.
  • the number of vehicles and pedestrians around the ego vehicle may reach dozens, and the computing power requirements for the autonomous driving tasks that the ego vehicle needs to handle will therefore reach a peak state.
  • processor cores can be further allocated to VD and PD tasks to ensure automatic driving.
  • the automatic driving system of the car can meet the computing power requirements of the vehicle detection around pedestrians and pedestrian detection tasks; in addition, the processor cores can also be fixedly assigned to the Router and MOP tasks to ensure the calculation of the path planning and motion control tasks of the self-vehicle. power needs can be met. Through this setting, it can be guaranteed that in such scenarios where the demand for computing power reaches a peak, the automatic driving task can be completed immediately and smoothly.
  • key targets around the vehicle can be selected, and the automatic driving system sets the perception and related planning of key targets as a higher priority Task
  • the strategy of key target selection can be determined according to the strength and weakness of road traffic participants. For example, if a vehicle collides with a pedestrian, it may cause serious injury consequences. Therefore, for example, pedestrian P1 in Figure 11 can be used as The key goal is that the automatic driving system tracks and predicts the trajectory of the pedestrian P1 with a higher priority, and makes corresponding trajectory planning for the vehicle.
  • non-motor vehicles such as bicycles
  • the automatic driving system can track and track non-motor vehicles with a higher priority. Trajectory prediction, and make corresponding vehicle trajectory planning.
  • the selection of key targets may include traffic participants who are more prone to accidents.
  • express delivery is a common type of work.
  • cities people engaged in catering express work
  • two-wheeled (motorized or electric) vehicles as food delivery tools, and are prone to traffic accidents. Therefore, in the selection of key targets, the two-wheeled vehicle on the road can be taken as the key target, such as the two-wheeled motor vehicle in Figure 11, the automatic driving system will track and predict the trajectory of the two-wheeled motor vehicle B1 with a higher priority, and Make corresponding vehicle trajectory planning to avoid traffic accidents.
  • the perception system of the autonomous driving system can be pre-trained. Specifically, if the perception system uses a neural network (such as a convolutional neural network) to process the acquired image information, then pre-training the neural network includes food delivery personnel wearing uniforms. After the training is completed (that is, the recognition accuracy of the neural network of the perception system for the food delivery person exceeds the preset value), the vehicle can use the perception system to identify the food delivery person, and take the delivery person as the key target to avoid traffic accidents with them.
  • a neural network such as a convolutional neural network
  • ResNet-50 can be used as the neural network used in the perception system. It should be understood that other types of neural networks (such as VGG) can also be used as the neural network used in the perception system, as long as it meets the requirements of It can be used within the spirit disclosed in this application.
  • VGG neural network used in the perception system
  • the food delivery staff may often set a red light or go against the road. Therefore, after the automatic driving system regards the food delivery person as a key target, it performs uninterrupted target tracking and trajectory prediction for the food delivery person in real time, and the automatic driving system plans the trajectory of the vehicle to avoid collision with the food delivery person.
  • the key target after the own vehicle determines each key target, the key target can also be highlighted/highlighted/emphasized displayed on the central control screen, so as to remind the driver of the own vehicle to pay attention to the above-mentioned key targets around the own vehicle.
  • the driver when the collision risk between the own vehicle and each key target exceeds a set threshold, the driver is reminded to take over the own vehicle.
  • the driver can be reminded acoustically, optically or in combination.
  • the self-vehicle By selecting key targets from road traffic participants and pre-training the perception system of the automatic driving system. It can make the self-vehicle avoid collision with VRU (Vulnerable Road Users: Vulnerable Road Users) as much as possible in the above-mentioned various complex traffic scenarios to cause personal injury. On the other hand, the self-vehicle can also avoid as much as possible Collisions with road traffic participants who are prone to traffic accidents (such as food delivery people). Therefore, the self-vehicle has higher automatic driving safety in the complex urban road environment.
  • VRU Vehicleable Road Users: Vulnerable Road Users
  • the selection of the above-mentioned key targets and the corresponding ego vehicle path planning are based on the premise of fixed allocation of processor cores to automatic driving tasks disclosed in this application.
  • the number of pedestrians and non-motorized vehicles on the road may be dozens or even hundreds, and the perception of key targets and their corresponding planning tasks can be given a higher task priority.
  • the task of perceiving and planning key targets can also be fixedly assigned to the processor core, so as to ensure the safety of the vehicle's automatic driving in the above-mentioned complex scenarios to the greatest extent possible.
  • the capability of the self-vehicle's perception system may be affected by large obstacles around it, for example, the self-vehicle's camera device may be blocked by large vehicles (such as buses, trucks, etc.) in front of the self-vehicle , so that the signal of the traffic light at the intersection ahead cannot be obtained.
  • the self-vehicle can obtain the signal of the front traffic light in real time through the interaction with the roadside equipment and based on the information interaction between the roadside equipment and the traffic light.
  • the automatic driving system when the self-vehicle is in various complex and difficult scenes, the automatic driving system will detect the traffic lights in the scene; if the automatic driving system determines that the traffic lights in the current scene are not working or degraded (For example, the aforementioned only shows yellow flash); then the automatic driving system limits the speed of the own vehicle, for example, the speed of the vehicle is limited below 40km/h.
  • the speed limit of the own vehicle when the traffic signal light is not working or degraded, the own vehicle can drive at a lower speed in such a scene, reducing the possibility of traffic accidents.
  • the above-mentioned automatic driving system determines that the traffic lights in the current scene are in different or degraded states, which can be achieved in a variety of ways;
  • the traffic lights maintain communication, and the roadside equipment sends the status of the traffic lights to the ego vehicle.
  • the perception system of the automatic driving system may include a neural network (such as a convolutional neural network), and the neural network is trained in advance, so that the neural network can recognize the state of traffic lights not working or degraded; and according to the trained neural network To determine whether the traffic lights are working.
  • the above-mentioned speed limit for the self-vehicle is still based on the fixed allocation of processor cores to the automatic driving task. Considering further, limiting the speed of the self-driving car will further help the automatic driving system to have sufficient time to process the acquired information and make plans, because a lower speed means that it takes more time to pass the same distance.
  • the speed limit of the self-vehicle can allow the automatic driving system to have more time to process information and make decisions in the above-mentioned difficult scenarios (traffic road conditions and traffic lights failure/degradation).
  • the automatic driving system adopts an interactive game strategy when sensing, tracking and making corresponding plans for the surrounding obstacles. Specifically: firstly, it is necessary to accurately identify the category and motion state information of the obstacles around the vehicle, and understand the intention of the obstacle based on the historical motion state of the obstacle and the road topology information, and then predict the possible behavior of the obstacle to achieve its intended goal Trajectory, obstacle interaction decision According to the possible behavior trajectory of the obstacle under different intentions and the behavior trajectory space of the self-vehicle, evaluate the possible interactive behavior trajectory space of the obstacle and the self-vehicle, and make an interactive decision.
  • the interactive game strategy adopted by the above-mentioned automatic driving system is also based on the fixed assignment of processor cores to the automatic driving tasks in this application.
  • the automatic driving system has enough computing power to process the perception and planning tasks of the surrounding dynamic obstacles in a timely manner.
  • the following describes the interactive game process between the ego vehicle and the surrounding dynamic obstacles.
  • the automatic driving system judges the part of the dynamic obstacles around the self-vehicle that belongs to the game target and the part that belongs to the non-game target. If there is a conflict between the future trajectory of the ego vehicle and the future trajectory of the dynamic obstacle, the dynamic obstacle is determined as the game target; otherwise, it is determined as the non-game target.
  • the automatic driving system of the own vehicle determines that there are two targets (or obstacles) within the current range according to the acquired perception information, which are car A and car B, and A Car and car B are in motion.
  • the automatic driving system can generate the predicted trajectory of car A and the predicted trajectory of car B according to the road topology and the motion state of the target.
  • the predicted trajectory of car B conflicts with the trajectory of the own vehicle, while the vehicle A is in the state of going straight in the same direction as the own vehicle, and the predicted trajectory of A vehicle does not conflict with the trajectory of the own vehicle.
  • the automatic driving system will B
  • the car is determined as the game target, that is, the self-car will play a game with the B car during the subsequent period of movement at the current moment, and the follow-up movement decision of the self-car needs to consider the predicted trajectory of the B car.
  • the perception module of the automatic driving system continuously observes and records the key information of the game target’s motion state, including horizontal state characteristics and vertical state characteristics, and the perception module sends this information to A planning module, the planning module generates a set of decision-making schemes based on the acquired information.
  • the planning module can generate a longitudinal acceleration sampling dimension based on the acquired perception information and considering longitudinal characteristic information such as road speed limit, acceleration change Jerk value, game target type, etc., considering road boundaries, static Obstacles, vehicle kinematics and other feature information generate lateral offset sampling dimensions; the planning module combines the lateral position offset sampling dimension with the longitudinal acceleration sampling dimension to form a sampling space, which is the possibility space of all strategies .
  • the planning module can obtain the set of all decision-making schemes according to the Zhangcheng sampling space.
  • the planning module may divide the decision-making solution set into at least one decision-making solution subset based on the category of the ego vehicle and the game target action pair included in each decision-making solution in the decision-making solution set. That is: the planning module can mark decision labels for all action pairs of the self-vehicle and the game target in the decision solution set, and divide the cluster with the same label into a decision solution subset, and each decision solution subset includes at least one decision solution.
  • the planning module labels each decision-making solution in the decision-making solution set with a decision label.
  • the decision labels in the decision scheme set may include, for example: Yield without avoidance, Preemption without avoidance, Preemptive and avoidance, and Yield and avoidance.
  • Each of the above labels represents the decision type of the ego car to deal with the game target.
  • the decision of "yield without avoidance” it means that the ego car will give way by decelerating in the subsequent game process, but it does not significantly change its own direction of movement (no avoidance);
  • the types of decisions represented by the above-mentioned decision tags can be quantitatively evaluated.
  • the income evaluation system for calculating the corresponding income of the decision-making scheme provided by the embodiment of the application can include the decision-making scheme There are several decision dimensions such as safety, comfort, passability, right of way, deviation, and historical decision results.
  • the decision-making benefits in the benefit evaluation system can be determined according to the decision-making cost (cost). The smaller the cost, the higher the decision-making benefits, and the higher the possibility that the decision-making scheme is the optimal decision-making scheme.
  • cost the decision-making cost
  • the specific evaluation dimensions and explanations of value are as follows:
  • the security cost in the embodiment of this application is mainly a decision cost value obtained based on the distance between the vehicle and the game target during the deduction of the trajectory of the vehicle and the trajectory of the game target.
  • a minimum distance between the self-vehicle and the game target can also be set.
  • the security cost gradually increases.
  • the security cost reaches the maximum, and the security-based decision-making profit reaches the minimum.
  • a maximum distance between the self-vehicle and the game target for calculating the safety cost can also be set.
  • the security cost Reaching the minimum the benefit of decision-making based on security reaches the maximum.
  • the horizontal axis represents the distance between the own car and the game target
  • the point A in the horizontal axis represents the distance between the own car and the game target set in the embodiment of the present application.
  • the minimum distance between, the point B in the horizontal axis can represent the maximum distance between the self-vehicle and the game target set in the embodiment of the present application for calculating the safety cost
  • the vertical axis represents the safety cost.
  • the security cost when the distance between the self-vehicle and the game target is not less than B, the security cost reaches the minimum, and the decision-making profit based on security reaches the maximum; when the distance between the self-vehicle and the game target is between point A and point B , the closer the distance between the self-vehicle and the game target is to point A, the greater the security cost is, and the smaller the profit of the decision-making based on security is, the closer the distance between the self-vehicle and the game target is to point B, the greater the security cost
  • the smaller the value is, the greater the benefit of security-based decision-making is; when the distance between the self-vehicle and the game target is not greater than A, the security cost reaches the maximum, and the security-based decision-making benefit reaches the minimum.
  • point A and point B shown in (a) in FIG. 15 in the embodiment of the present application may be obtained based on actual experience.
  • point A shown in (a) in Figure 15 in the embodiment of the present application may be 0.2m to 0.5m (for example, point A is 0.5m)
  • point B may be 1.0m to 3.0m (for example, point B is 2.5m).
  • Comfort cost is mainly a decision cost value obtained based on the difference between the current acceleration of the ego vehicle and the expected acceleration.
  • a maximum difference between the current acceleration of the ego vehicle and the expected acceleration can also be set.
  • the difference between the current acceleration of the ego vehicle and the expected acceleration is closer to the maximum difference, the greater the comfort cost, Comfort-based decision-making benefits are smaller.
  • the horizontal axis represents the difference between the current acceleration of the ego vehicle and the expected acceleration, for example, the current acceleration of the ego vehicle is 1m/S ⁇ 2, and the expected acceleration is 4m /S ⁇ 2, the difference between the current acceleration of the ego vehicle and the expected acceleration is 3m/S ⁇ 2.
  • Point A on the horizontal axis represents the maximum difference between the current acceleration of the ego vehicle and the expected acceleration set in the embodiment of the present application.
  • the comfort cost when the difference between the current acceleration of the ego vehicle and the expected acceleration is closer to point A, the greater the comfort cost, the smaller the benefit of decision-making based on comfort; when the difference between the current acceleration and the expected acceleration of the ego vehicle is farther away from point A , the smaller the comfort cost, the greater the benefit of decision-making based on comfort; when the difference between the current acceleration of the ego vehicle and the expected acceleration is not less than the acceleration value corresponding to point A, the comfort cost reaches the maximum, and the benefit of decision-making based on comfort reach the minimum.
  • the point A shown in (b) in FIG. 15 in the embodiment of the present application may be obtained based on actual experience.
  • point A shown in (b) in FIG. 15 in the embodiment of the present application may be 5m/S ⁇ 2 ⁇ 7m/S ⁇ 2 (for example, point A is 6m/S ⁇ 2).
  • the passability cost in the embodiment of this application is mainly based on the speed difference between the current speed of the ego vehicle and the speed at which the ego vehicle reaches the conflict point (for example, the point where the ego vehicle and the game target trajectory may interact) Worth getting a decision cost value.
  • a maximum value of the speed difference may also be set.
  • the passability cost is greater, and the passability-based decision-making benefit is smaller.
  • the horizontal axis represents the speed difference
  • point A on the horizontal axis represents the maximum value of the speed difference set in the embodiment of the present application.
  • the passability cost is greater, and the decision-making benefit based on passability is smaller; when the speed difference is farther away from point A, the passability cost is smaller, and passability-based decision-making benefits are smaller.
  • the greater the decision-making benefit when the speed difference is not less than the speed difference corresponding to point A, the passability cost reaches the maximum, and the passability-based decision-making profit reaches the minimum.
  • the point A shown in (c) in FIG. 15 in the embodiment of the present application may be obtained based on actual experience.
  • point A shown in (c) in FIG. 15 in the embodiment of the present application may be 6m/s ⁇ 10m/s (for example, point A is 7m/s).
  • Right-of-way cost is a decision cost value obtained for the current acceleration of the own vehicle based on the right-of-way situation of the own vehicle or the game target.
  • the calculation is mainly based on the party with the high right of way, and the right of way cost penalty is imposed on the party with the right of way. For example, if the own vehicle has the high right of way, then if the own vehicle slows down, then Right cost penalty.
  • the right of way in the embodiment of the present application is the relative relationship between the self-vehicle and the game target.
  • the car with the right of way should be kept from changing its motion state as much as possible. Therefore, if the acceleration of the self-vehicle decreases and the vehicle slows down, the change of the motion state of the self-vehicle should be punished with a higher cost of the road right.
  • a minimum value of the current acceleration of the ego vehicle can also be set.
  • the cost of the right of way is greater, and the benefit of decision-making based on the right of way is smaller.
  • ACC represents acceleration, where the right of way of the own vehicle is a limited premise, for example, the right of way of the own vehicle is input as a premise, based on the vehicle and the game Target the current right-of-way relationship, and further determine the right-of-way cost.
  • the horizontal axis represents the current acceleration of the ego vehicle
  • point A on the abscissa represents the minimum difference of the current acceleration of the ego vehicle set in the embodiment of the present application.
  • the decision-making income when the current acceleration of the ego vehicle is closer to point A, the greater the cost of the right of way, the smaller the benefit of decision-making based on the right of way; The greater the decision-making income; when the current acceleration of the ego vehicle is not greater than the acceleration value corresponding to point A, the cost of the right of way reaches the maximum, and the decision-making income based on the right of way reaches the minimum.
  • the point A shown in (d) in FIG. 15 in the embodiment of the present application may be obtained based on actual experience.
  • the point A shown in (d) in Figure 15 in the embodiment of the present application may be (-1)m/S ⁇ 2 ⁇ (-3)m/S ⁇ 2 (for example, point A is (-1) m/S ⁇ 2).
  • Offset cost is mainly a decision cost value obtained based on the position offset of the ego vehicle relative to the reference line. Among them, it can be understood that the farther the offset distance of the ego vehicle relative to the reference line is, the larger the offset cost value is, and the smaller the benefit of the ego vehicle based on the offset decision is.
  • the reference line in this embodiment of the present application may be the centerline of the road where the vehicle is located.
  • the embodiment of the present application may include two offset regions, such as region A and region B.
  • region A may be called a soft boundary
  • region B may be called a hard boundary.
  • the soft boundary can be the area of the driving lane where the vehicle is located, that is to say, if the ego vehicle is offset relative to the reference line, but the ego vehicle has been driving in the driving lane, it can be understood that the ego vehicle has been in the soft boundary; this application
  • the hard boundary set in the embodiment may be an area beyond the driving lane, that is, if the ego vehicle deviates from the reference line and the ego vehicle leaves the driving lane, it can be understood that the ego vehicle has entered the hard boundary.
  • the embodiments of the present application may use different slope values for calculating the offset cost based on different regions, where the slope value for calculating the offset cost set in the soft boundary can be smaller, and set in the hard boundary The slope value used to calculate the offset cost can be larger.
  • the horizontal axis represents the offset distance from the current reference line of the ego vehicle
  • the -A point in the horizontal axis represents the maximum soft boundary set in the embodiment of the present application.
  • the position on the left side, point 0 indicates the position of the reference line
  • point A indicates the rightmost position of the soft boundary set in the embodiment of the present application
  • point -B indicates the leftmost position of the hard boundary set in the embodiment of the application
  • point B Indicates the rightmost position of the hard border set in the embodiment of this application.
  • the offset cost of the vehicle relative to the reference line in the soft boundary is calculated using slope 1
  • the offset cost of the vehicle relative to the reference line in the hard boundary is calculated using slope 2
  • slope 2 is greater than slope 1.
  • the offset cost is larger, and the decision-making benefit based on the offset is smaller.
  • the offset cost reaches the maximum, and the benefit of the decision based on the offset reaches the minimum.
  • point A, point B, and slope 1 and slope 2 shown in (e) in FIG. 9-4 in the embodiment of the present application may be obtained based on actual experience.
  • point A shown in (e) in Figure 9-4 in the embodiment of the present application may be 1.5m to 2.5m (for example, if point A is 2m, then the corresponding point -A is -2m), and point B may be 1.9m ⁇ 2.9m (for example, point B is 2.9m, then the corresponding -B point is -2.9m).
  • Historical decision result cost in the embodiment of this application can also be called the inter-frame correlation cost, which is mainly based on the decision scheme of the last frame of the ego vehicle, and a decision cost based on the current frame.
  • the electronic device in the embodiment of the present application may fuse the above-mentioned costs of each evaluation dimension according to a certain weight ratio, so as to obtain a multi-dimensional decision cost value.
  • the profit evaluation system may be mainly determined by historical decision results, that is, it may be mainly based on historical decision results (for example, setting the weight of historical decision results cost at 80%-90%), and determined according to historical decision results The driving style of the vehicle, eg conservative, moderate or aggressive.
  • the embodiments of the present application set the weights corresponding to each evaluation dimension in the income evaluation system as:
  • Safety is 10,000, passability is 6,000, comfort is 500, right of way is 1,000, offset is 2,000, and risk area is 1,000.
  • the decision cost value obtained in the embodiment of this application can be:
  • the weights corresponding to each evaluation dimension in the revenue evaluation system may be different.
  • the corresponding relationship between the driving environment and the weight of the evaluation dimension may be shown in Table 1 below.
  • the strategic feasible region generation module can determine the weights corresponding to each evaluation dimension in the revenue evaluation system adopted according to the contents of Table 1 as safety cost*10000, passability cost*6000, Comfort cost*500, right of way cost*1000, offset cost*2000, risk area cost*1000.
  • the decision-making benefit value of the self-vehicle can be obtained.
  • the trajectories of the ego vehicle and the game target corresponding to different specific decision-making schemes are illustrated by dotted lines and solid lines. Taking "Give way without avoidance" as an example, under this decision-making scheme, the ego vehicle and the game target each have three corresponding trajectories.
  • the game target corresponds to the one with the largest benefit value of the self-vehicle's decision-making in the trajectory (that is, in the current strategy evaluation of yielding without avoidance, it is the most "favorable" for the ego-vehicle to drive according to the planned trajectory), correspondingly, the game target
  • the track marked by the solid line represents the predicted track of the game target corresponding to the track of the ego vehicle.
  • the above-mentioned solid-line trajectory of the ego vehicle is called the target decision-making scheme in the subset of decision-making schemes of "give way and no avoidance". Similarly, target decision scenarios for other subsets of decision scenarios can be determined.
  • trajectory pairs self-vehicle, game target
  • yield without avoidance the number of trajectory pairs can also be other natural numbers, such as four trajectory pairs, or two trajectories are equivalent.
  • the target decision-making scheme in each subset of decision-making schemes can be determined.
  • the decision-making scheme subsets can be divided into different echelons. Still taking FIG. 14 as an example, four decision-making scheme subsets are shown : Yield without evasion, preemption without evasion, preemptive evasion, evasion of evasion. Each subset of decision scenarios has a target decision scenario. By comparing the decision profit value of the target decision-making schemes of the above four different decision-making scheme subsets, the above-mentioned four different decision-making scheme subsets can be divided into different "echelons" according to a certain threshold.
  • the threshold is set to 0.76 (threshold can be obtained according to experience value or historical value)
  • the decision-making profit values of preemptive avoidance and yielding without avoidance are respectively 0.8 and 0.77, they are all greater than the threshold; while yield without avoidance and yield with avoidance are 0.73 and 0.69, they are all less than the set threshold.
  • the decision-making schemes of preemption and avoidance and yield without avoidance can be regarded as the first echelon
  • preemption without avoidance and yield and avoidance can be regarded as the second echelon.
  • the first echelon represents a better "strategy" than the second echelon.
  • this strategy can not only meet the needs of the self-vehicle to pass smoothly, but also ensure driving safety.
  • It is said to be a "pretty good” strategy; and if the self-vehicle chooses to give way without avoiding, driving safety can be guaranteed, and the self-vehicle does not need to perform lateral avoidance.
  • This strategy ensures safety, and it is generally a "quite good” strategy " strategy. For the strategy of rushing without avoidance, although this can ensure the smoothness of the own vehicle, but because there is no avoidance action in the lateral direction, the probability of collision between the straight-going own vehicle and the incoming other vehicle is relatively high.
  • the strategy sets of the first echelon and the second echelon mentioned above refer to the current state of the self-vehicle. As the states of the self-vehicle and the game target change, the profit scores and Corresponding priorities are subject to change at any time.
  • the planning module further confirms the decision-making benefit difference between the target decision-making plans of the first echelon , when the difference between the decision-making benefits is less than the set threshold, it is considered that there is no optimal plan among the target decision-making plan sets of the first echelon.
  • the above-mentioned first echelon includes two target decision-making schemes of rushing and yielding and yielding without avoiding, and their decision-making returns are 0.8 and 0.77 respectively.
  • the threshold is set to 0.1
  • the two target decision-making schemes If the profit difference (0.03) between the two target decision-making schemes is less than the set threshold, then the planning module believes that there is no obvious difference between the above two target decision-making schemes at the current moment.
  • Whichever scheme can be adopted it can be understood that the interaction relationship between the self-vehicle and the game target is not clear at this time, and it is necessary to enter the first interaction stage, in which the self-vehicle and the game target are kept corresponding Possibility to jump between multiple different decision-making options.
  • the planning module determines that the interaction between the self-vehicle and the game target satisfies the conditions for entering the second interaction stage, it enters the second interaction stage from the first interaction stage. At this time, the interaction relationship between the self-vehicle and the game target is clear.
  • the planning module may judge whether the ego vehicle enters the second interaction stage from the first interaction stage based on the following conditions.
  • Condition 1 The planning module can determine whether the entry is satisfied based on the size relationship between the decision benefit difference between the target decision plan with the highest decision benefit and the target decision plan with the second highest decision benefit in the set of target decision plans and the threshold benefit difference. Conditions for the second interaction phase. When the decision benefit difference between the decision plan with the highest decision benefit and the decision plan with the second highest decision benefit in the set of decision plans is greater than the threshold benefit difference, the planning module determines that the condition for entering the second interaction stage is satisfied.
  • the planning module can determine whether the condition for entering the second interaction stage is satisfied based on the comparison result between the actual trajectory of the game object at the predetermined moment and the trajectory of the game object predicted by the first decision-making scheme.
  • the planning module determines that the actual trajectory of the game object at the predetermined moment is relatively consistent with the trajectory of the game object predicted by the first decision-making scheme at the predetermined moment, the planning module determines that it is satisfied and enters the second The condition of the interaction phase.
  • the first decision-making scheme adopted in the first interaction stage is the same as the second decision-making scheme adopted in the second interaction stage.
  • condition 2 when the planning module determines that the actual trajectory of the game object at the predetermined moment is completely inconsistent with the trajectory of the game object predicted by the first decision-making scheme at the predetermined moment, the planning module determines that it is satisfied and enters the second interaction stage conditions. It can be understood that when the planning module determines that the actual trajectory of the game target at the predetermined moment is completely inconsistent with the trajectory of the game target at the predetermined moment predicted by the target decision-making scheme (for example: the above-mentioned "preemption and avoidance"), it means that the game target's The trajectory is completely unexpected, which means that the current decision-making scheme for game interaction must not be applicable.
  • the target decision-making scheme for example: the above-mentioned "preemption and avoidance
  • the planning module needs to adjust the current target decision-making scheme A for game interaction to a more suitable target decision-making scheme for this game ( For example, the target decision-making scheme B), the decision-making scheme B can be obtained from the set of target decision-making schemes, for example, if the current self-vehicle finds that the rushing avoidance strategy is no longer applicable (the game target may not give way as expected but rush ahead without deceleration ), then a more suitable target decision-making scheme can be obtained in the decision-making scheme set, such as yielding and avoiding.
  • the planning module can confirm that the current interactive relationship between the self-vehicle and the game target is clear, and enter the second interaction stage.
  • the planning module can also combine condition one and condition two to jointly determine whether the condition for entering the second interaction stage is satisfied.
  • condition 1 and condition 2 When combining condition 1 and condition 2 to determine whether the conditions for entering the second interaction stage are satisfied, if the result of judging whether the condition is satisfied and entering the second interaction stage based on condition 1 is the same as judging whether the condition for entering the second interaction stage is satisfied based on condition 2
  • the priority of condition 2 is higher than that of condition 1.
  • the reason for this setting is: need to pay attention to the consistency/inconsistency Jumping is usually due to the fact that the game target does not follow the traffic rules or general driving habits, which may be contrary to the evaluation of the cost, and its priority is higher than jumping based on revenue.
  • the planning module can judge consistency or inconsistency according to the horizontal and vertical action sequences of the game object.
  • the planning module when the planning module determines whether the predicted trajectory at the predetermined moment of the game target is consistent with the real trajectory at the predetermined moment of the game target, the planning module can use the continuous Judgment is based on historical state changes, based on the action sequence predicted by the game target in the interactive decision-making scheme selected at each time and the actual action sequence of the game target. In the process of judgment, horizontal and vertical can be distinguished.
  • the expected acceleration and the actual acceleration sequence are processed using a statistical data processing method to obtain a longitudinal action sequence as shown in (a) in FIG. 17 .
  • a1 ⁇ a3 are the expected acceleration changes of the game target within the expected duration determined by the planning module
  • b1 ⁇ b3 are the actual acceleration of the game target within the current duration
  • b4-b6 is the actual acceleration change of the game target within the expected time period.
  • the planning module determines the tracking result of the game target at the predetermined time and The trajectory of the game object predicted by the first decision-making scheme at the predetermined time is inconsistent; if the difference between a1-a3 and b4-b6 fluctuates within the threshold range, the planning module determines that the tracking result of the game object at the predetermined time is consistent with the first decision-making scheme The trajectory of the predicted game target at the predetermined moment is consistent.
  • the sequence of expected distance and actual distance is processed by using the statistical data processing method to obtain the sequence of lateral actions as shown in (b) in FIG. 17 .
  • the judgment of horizontal consistency/inconsistency will be affected by the historical judgment results.
  • the planning module determines that the tracking result of the game target at the predetermined time is consistent with the The trajectory of the game object predicted by the first decision-making scheme at the predetermined moment is inconsistent; if the fluctuation between c1 ⁇ c3 and d4 ⁇ d6 is within the threshold range, the planning module determines that the tracking result of the game object at the predetermined moment is consistent with the first The trajectory of the game target predicted by the decision-making scheme is consistent at the predetermined moment.
  • this application can divide the interactive game process into multiple interactive stages according to the actual situation.
  • the first interaction stage may be called a critical state.
  • the critical state can be understood as the interaction relationship between the self-vehicle and the game target is not clear, that is, the decision-making scheme for the interaction between the self-vehicle and the game target may change at any time.
  • the second interaction stage may be called a confidence state.
  • the confidence state can be understood as the clear interaction relationship between the self-vehicle and the game target, that is, the determination of the decision-making scheme for the interaction between the self-vehicle and the game target.
  • the self-vehicle will approach a decision-making plan.
  • the difference between the decision-making income of the approaching decision-making plan and the decision-making income of other decision-making plans is greater
  • the threshold returns are different, it is considered that the game relationship is clear, and the jump from the critical state to the confidence state is completed.
  • Fig. 18 shows the overall process logic of the above-mentioned interactive game.
  • the device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • the connection relationship between the modules indicates that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines.
  • the essence of the technical solution of this application or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product is stored in a readable storage medium, such as a floppy disk of a computer , U disk, mobile hard disk, ROM, RAM, magnetic disk or optical disk, etc., including several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application .
  • a computer device which can be a personal computer, a server, or a network device, etc.
  • all or part of them may be implemented by software, hardware, firmware or any combination thereof.
  • software When implemented using software, it may be implemented in whole or in part in the form of a computer program product.
  • the computer program product includes one or more computer instructions.
  • the computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website, computer, server, or data center Transmission to another website site, computer, server, or data center by wired (eg, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (eg, infrared, wireless, microwave, etc.).
  • wired eg, coaxial cable, optical fiber, digital subscriber line (DSL)
  • wireless eg, infrared, wireless, microwave, etc.
  • the computer-readable storage medium may be any available medium that can be stored by a computer, or a data storage device such as a server or a data center integrated with one or more available media.
  • the available medium may be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (Solid State Disk, SSD)), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Mechanical Engineering (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Transportation (AREA)
  • Human Computer Interaction (AREA)
  • General Engineering & Computer Science (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种自动驾驶装置，包括：处理装置，处理装置包括多个处理器核；多个处理器核中的至少一部分处理器核被固定分配给设定的任务集合，固定分配指：多个处理器核中的至少一部分处理器核被配置为仅处理设定的任务集合，设定的任务集合为非空集合。通过预先将自动驾驶系统的处理器核中的大多数固定分配给自动驾驶任务，可以保证即使在多目标（障碍物）的场景下，处理器核所提供的算力足可以满足在各种场景下的自动驾驶任务中的感知和规划任务，从而避免了因为不能即时的处理感知和规划任务所导致的无法实时地给出自动驾驶策略的问题。

Description

一种智能驾驶方法、装置及包括该装置的车辆 技术领域
本申请涉及智能驾驶技术领域，尤其涉及一种基于智能驾驶方法、装置及包括该装置的车辆。
背景技术
近年来，智能驾驶技术在快速发展，智能驾驶又可以称为自动驾驶或辅助驾驶，是车辆智能化发展的重要方向，随着感知技术的发展以及芯片能力的提升，智能驾驶为人们提供了越来越多的丰富的驾驶功能，逐渐实现不同级别的驾驶体验。自动机工程师学会(society of automotive engineers，SAE)提供了一种驾驶自动化分级标准，包括驾驶等级L0至L5，其中L0级为无自动化，由人类驾驶者全权操作汽车，在行驶过程中可以得到驾驶系统的警告或辅助，例如自动紧急制动(autonomous emergency braking，AEB)，盲点检测(blind spot monitoring，BSM)或车道偏离报警(lane departure warning，LDW)等。L1级为驾驶支援，驾驶操作由人类驾驶者和驾驶系统共同完成，驾驶系统可以通过驾驶环境对方向盘或加减速操作提供驾驶支援，其他的驾驶操作由人类驾驶员进行，例如自适应巡航控制(adaptive cruise control，ACC)或车道保持辅助/支持(lane keep assistance/support，LKA/LKS)等；L2级为部分自动化，通过驾驶环境对方向盘和加减速中的多项提供驾驶支援，其他的驾驶动作由人类驾驶员进行，例如结合了自适应巡航控制(adaptive cruise control，ACC)和车道保持辅助(lane keep assistance，LKA)的跟车功能；L3级为有条件自动化，可以由驾驶系统完成所有的驾驶操作，但人类驾驶员需要在适当的时候应答驾驶系统的请求，即人类驾驶员需要做好接管驾驶系统的准备；L4级为高度自动化，可以由驾驶系统完成所有的驾驶操作，人类驾驶员不一定需要对驾驶系统的请求做出应答，例如在道路和环境条件允许的情况下(比如封闭的园区、高速公路、城市道路或固定的行车线路等)人类驾驶员可以不接管驾驶；L5级为完全自动化，在各种人类驾驶员可以应对的道路和环境条件下的驾驶操作均可以由驾驶系统自主完成。可见，L0至L2的级别，驾驶系统主要为驾驶员提供支持，驾驶员仍然需要做好驾驶监督，根据需要进行转向、制动或加速以保证安全。L3至L5级别，驾驶系统可以代替驾驶员完成所有的驾驶操作，L3级别下，驾驶员要做好接管驾驶的准备，L4和L5级别驾驶系统可以实现部分条件和所有条件下的完全驾驶，驾驶员可以选择是否接管。
目前，对于具有智能驾驶能力的车辆(一般指上述的SAE分级在L2及以上的车辆)而言，在进行自动驾驶的时候，一个普遍而困难的问题在于：当车辆处道路的左转、调头或环岛场景时，车辆普遍难以处理这些场景下的自动驾驶任务。原因主要在于：(1)在道路的左转场景，尤其是无保护左转(即路口无转向车道导引线)的场景下，发生在各类平交或立交路口，交通流密度高，障碍物多，车辆在执行自动驾驶任务的时候，难以处理和多个障碍物的博弈关系；(2)由于路口的需要识别的目标(障碍物) 不仅数量多，而且种类也较多，例如，车辆在路口需要识别的静态目标包括：车道线、人行道、停止线、导向箭头牌、限速标识等；而需要识别的动态目标包括：公交车、卡车、乘用车、两轮车、行人等，这些动态目标之间往往还存在混行的情况(例如行人和车辆同时出现在斑马线上)，车辆的自动驾驶系统需要不停地追踪这些动态障碍物的运动情况以不断地调整自动驾驶策略。基于以上的原因，在左转场景下(尤其是城区的左转场景)，整个道路体系场景(路口)属于非结构化道路，路权冲突和竞争关系复杂多变，需要规避的障碍物数目众多、状态多变，这将使得车辆的自动驾驶系统对算力资源的需求产生瞬时高峰，这种对算力的瞬时需求高峰会使得自动驾驶系统无法及时地处理当前的信息，导致无法给出实时可用的自动驾驶策略，并最终可能导致交通事故的发生。
进一步地，在其它的一些道路场景下(例如路口右转、或者环岛)，往往也存在上述目标众多、路权冲突和竞争关系复杂多变的情况，因此往往也会对车辆的自动驾驶系统产生瞬时算力高峰需求，并导致和上述问题相类似的问题。
综上，需要一种智能驾驶方法，其可以使得车辆的自动驾驶系统可以平稳、可靠地处理多个动态目标跟踪的道路场景(尤其是左转场景)，提升车辆的自动驾驶的安全性。
发明内容
在一些实施例中，提供一种自动驾驶装置，自动驾驶装置可以用于具有智能驾驶能力的车辆(例如上述的L2-L5级自动驾驶能力的车辆)。自动驾驶装置包括：处理装置，处理装置包括多个处理器核；多个处理器核中的至少一部分处理器核被固定分配给设定的任务集合，所述固定分配指：所述多个处理器核中的至少一部分处理器核被配置为仅处理设定的任务集合，所述设定的任务集合为非空集合。通过预先将自动驾驶系统的处理器核中的大多数固定分配给自动驾驶任务，可以保证即使在多目标(障碍物)的场景下，处理器核所提供的算力足可以满足在上述各种场景下的自动驾驶任务中的感知和规划任务。从而避免了因为不能即时的处理感知和规划任务所导致的无法实时地给出自动驾驶策略的问题。
在一些实施例中，自动驾驶装置还包括用于获取自车感知信息的感知系统，感知系统包括以下至少一种：激光雷达，毫米波雷达，摄像头，超声波雷达，在具有智能驾驶能力的车辆上，可以设置多个同一类型的传感器，例如，可以在车辆上设置两个或者三个激光雷达。设定的任务集合包括感知任务，所述感知任务包括由所述至少一部分处理器核处理由所述感知系统所获取的信息并获取车辆周围的环境信息和障碍物信息。
在一些实施例中，障碍物信息包括以下至少一种：静态障碍物信息，动态障碍物信息。静态障碍物可以包括例如道路边的房屋、路侧停放的车辆、路灯柱、垃圾桶等；动态障碍物可以包括例如行驶中的车辆、行人等。
在一些实施例中，自动驾驶装置还包括规划系统；所述设定的任务集合包括规划任务，所述规划任务包括由所述的至少一部分处理器核基于所述车辆的运动信息核所述障碍物信息确定所述车辆的规划行驶策略。
在一些实施例中，感知任务包括以下至少一种：用于识别车辆周围车道线的车道线识别(LD)任务，用于识别车辆周围其它车辆的车辆检测(VD)，用于将不同传感器信息进行融合的感知融合(RL)，对车辆周围的关键目标进行感知和跟踪，所述关键目标包括弱势道路使用者(VRU)；所述规划任务包括以下至少一种：车辆的全局路径规划(Router)，有限状态机(FSM)，运动控制(MOP)。
在一些实施例中，将所述至少一部分处理器核中的至少一个处理器核固定分配给所述感知任务中的一种任务。和/或将所述至少一部分处理器核中的至少一个处理器核固定分配给所述规划任务中的一种任务。在自动驾驶任务中，主要包括感知任务和规划任务，规划系统基于感知系统所获取的信息对车辆的形式策略进行规划，通过将多个处理器核中的至少一部分固定分配给感知任务和/或规划任务，可以保证车辆自动驾驶过程中规划任务的顺利完成。
在一些实施例中，将至少一部分处理器核中的至少一个处理器核固定分配给对车辆周围的关键目标进行感知和跟踪，所述关键目标包括弱势道路使用者(VRU)，在车辆进行自动驾驶的过程中，安全是第一优先级任务，通过将对车辆周围的关键目标进行感知和跟踪的任务固定于至少一个处理器核，可以确保在复杂的场景下，车辆可以持续连贯地对车辆周围的关键目标，特别是例如行人的弱势道路使用者的跟踪，从而可以防止出现因为跟踪中断而引发的安全事故。
在一些实施例中，自动驾驶装置根据感知信息获取自车的决策方案集；并从所述决策方案集中获取用于所述自车与博弈目标进行第一交互阶段的第一决策方案；在基于所述第一决策方案控制所述自车进行行驶的过程中，确定所述自车与所述博弈目标满足进入第二交互阶段的条件；从所述决策方案集中获取用于所述自车与所述博弈目标进行第二交互阶段的第二决策方案；基于所述第二决策方案控制所述自车进行行驶。通过交互博弈的策略，可以使得车辆实时地依据周围的情况做出合适的自动驾驶策略。
在一些实施例中，所述第一交互阶段为自车与所述博弈目标的交互关系不明确的阶段；所述第二交互阶段为自车于所述博弈目标的交互关系明确的阶段。通过将车辆和周围环境的状况分为交互关系不明确阶段和交互关系明确阶段，来确定不同的车辆和周围环境的博弈策略。在交互关系不明确的阶段，车辆可以在多种交互策略中进行选择和切换(因为当前没有收益最优的策略)，在交互关系明确的阶段，车辆一般会有收益最优的策略。
在一些实施例中，自动驾驶装置配置为：根据所述感知信息确定至少一个目标；获取所述至少一个目标的未来运动轨迹；获取所述自车的未来运动轨迹；将所述至少一个目标中未来运动轨迹与所述自车的未来运动轨迹有交叉的目标确定为所述博弈目标，在所述自车与所述博弈目标进行所述第一交互阶段的过程中，对所述博弈目标的状态进行持续跟踪。通过对车辆周围的目标(障碍物)进行博弈目标和非博弈目标的区分，使得车辆可以持续关注博弈目标(即在未来可能会和车辆有冲突的目标)，从而可以减少甚至不持续地分配处理器核的算力在非博弈目标上(即在未来和车辆不会有冲突的目标)，从而可以节约算力。
在一些实施例中，包括一种智能驾驶车辆，其包括上述的各种实施例中的自动驾驶装置。
附图说明
图1示出了智能驾驶车辆的构成示意图；
图2示出了根据本申请实施例的自动驾驶系统的框架示意图；
图3-1示出了自动驾驶中一种常见的场景示意图；
图3-2示出了自动驾驶中一种常见的场景示意图；
图4示出了某自动驾驶车辆在实际场景下一段时间内所跟踪的目标数量；
图5示出了自动驾驶中一种常见的场景示意图；
图6示出了自动驾驶中一种常见的场景示意图；
图7示出了自动驾驶中一种常见的场景示意图；
图8示出了自动驾驶中一种常见的场景示意图；
图9示出了自动驾驶中一种常见的场景示意图；
图10示出了一种自动驾驶系统的处理器核的示意图；
图11示出了自动驾驶中一种常见的场景示意图；
图12示出了根据本申请实施例确定自车博弈目标的示意图；
图13示出了根据本申请实施例确定自车决策方案集的原理示意图；
图14示出了根据本申请实施例对自车决策方案集进行标注的示意图；
图15示出了根据本申请实施例对自车的决策方案进行收益评估的评估维度示意图；
图16示出了根据本申请实施例的偏移区间的示意图；
图17示出了根据本申请实施例的对自车动作序列进行处理的示意图；
图18示出了根据本申请实施例的自车和博弈目标的交互博弈的流程示意图。
具体实施方式
以下将参考附图详细说明本申请的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
下面首先对自动驾驶车辆(或称智能驾驶车辆)和系统进行示意性说明。应当理解的是：在本申请后续部分所涉及到的“自动驾驶”或“自动驾驶车辆”，指按照SAE分级在L2及以上能力的车辆。
图1为车辆100的一个功能框图示意。可以将车辆100配置为完全或部分自动驾驶模式。例如：车辆100可以通过感知系统120获取其周围的环境信息，并基于对周边环境信息的分析得到自动驾驶策略以实现完全自动驾驶，或者将分析结果呈现给用户以实现部分自动驾驶。
车辆100可包括各种子系统，例如信息娱乐系统110、感知系统120、决策控制系统130、驱动系统140以及计算平台150。可选地，车辆100可包括更多或更少的子系统，并且每个子系统都可包括多个部件。另外，车辆100的每个子系统和部件可以通 过有线或者无线的方式实现互连。
在一些实施例中，信息娱乐系统110可以包括通信系统111，娱乐系统112以及导航系统113。
通信系统111可以包括无线通信系统，无线通信系统可以直接地或者经由通信网络来与一个或多个设备无线通信。例如，无线通信系统146可使用3G蜂窝通信，例如CDMA、EVD0、GSM/GPRS，或者4G蜂窝通信，例如LTE。或者5G蜂窝通信。无线通信系统可利用WiFi与无线局域网(wireless local area network，WLAN)通信。在一些实施例中，无线通信系统146可利用红外链路、蓝牙或ZigBee与设备直接通信。其他无线协议，例如各种车辆通信系统，例如，无线通信系统可包括一个或多个专用短程通信(dedicated short range communications，DSRC)设备，这些设备可包括车辆和/或路边台站之间的公共和/或私有数据通信。
娱乐系统112可以包括中控屏，麦克风和音响，用户可以基于娱乐系统在车内收听广播，播放音乐；或者将手机和车辆联通，在中控屏上实现手机的投屏，中控屏可以为触控式，用户可以通过触摸屏幕进行操作。在一些情况下，可以通过麦克风获取用户的语音信号，并依据对用户的语音信号的分析实现用户对车辆100的某些控制，例如调节车内温度等。在另一些情况下，可以通过音响向用户播放音乐。
导航系统113可以包括由地图供应商所提供的地图服务，从而为车辆100提供行驶路线的导航，导航系统113可以和车辆的全球定位系统121、惯性测量单元122配合使用。地图供应商所提供的地图服务可以为二维地图，也可以是高精地图。
感知系统120可包括感测关于车辆100周边的环境的信息的若干种传感器。例如，感知系统120可包括全球定位系统121(全球定位系统可以是GPS系统，也可以是北斗系统或者其他定位系统)、惯性测量单元(inertial measurement unit，IMU)122、激光雷达123、毫米波雷达124、超声雷达125以及摄像装置126。感知系统120还可包括被监视车辆100的内部系统的传感器(例如，车内空气质量监测器、燃油量表、机油温度表等)。来自这些传感器中的一个或多个的传感器数据可用于检测对象及其相应特性(位置、形状、方向、速度等)。这种检测和识别是车辆100的安全操作的关键功能。
全球定位系统121可用于估计车辆100的地理位置。
惯性测量单元122用于基于惯性加速度来感测车辆100的位置和朝向变化。在一些实施例中，惯性测量单元122可以是加速度计和陀螺仪的组合。
激光雷达123可利用激光来感测车辆100所位于的环境中的物体。在一些实施例中，激光雷达123可包括一个或多个激光源、激光扫描器以及一个或多个检测器，以及其他系统组件。
毫米波雷达124可利用无线电信号来感测车辆100的周边环境内的物体。在一些实施例中，除了感测物体以外，雷达126还可用于感测物体的速度和/或前进方向。
超声雷达125可以利用超声波信号来感测车辆100周围的物体。
摄像装置126可用于捕捉车辆100的周边环境的图像信息。摄像装置126可以包括单目相机、双目相机、结构光相机以及全景相机等，摄像装置126获取的图像信息可以包括静态图像，也可以包括视频流信息。
决策控制系统130包括基于感知系统120所获取的信息进行分析决策的计算系统131，决策控制系统130还包括对车辆100的动力系统进行控制的整车控制器132，以及用于控制车辆100的转向系统133、油门134(包括电动车的加速踏板，这里是一个在一些实施例中称呼)和制动系统135
计算系统131可以操作来处理和分析由感知系统120所获取的各种信息以便识别车辆100周边环境中的目标、物体和/或特征。所述目标可以包括行人或者动物，所述物体和/或特征可包括交通信号、道路边界和障碍物。计算系统131可使用物体识别算法、运动中恢复结构(Structure from Motion，SFM)算法、视频跟踪等技术。在一些实施例中，计算系统131可以用于为环境绘制地图、跟踪物体、估计物体的速度等等。计算系统131可以将所获取的各种信息进行分析并得出对车辆的控制策略。
整车控制器132可以用于对车辆的动力电池和引擎141进行协调控制，以提升车辆100的动力性能。
转向系统133可操作来调整车辆100的前进方向。例如在一个实施例中可以为方向盘系统。
油门134用于控制引擎141的操作速度并进而控制车辆100的速度。
制动系统135用于控制车辆100减速。制动系统135可使用摩擦力来减慢车轮144。在一些实施例中，制动系统135可将车轮144的动能转换为电流。制动系统135也可采取其他形式来减慢车轮144转速从而控制车辆100的速度。
驱动系统140可包括为车辆100提供动力运动的组件。在一个实施例中，驱动系统140可包括引擎141、能量源142、传动系统143和车轮144。引擎141可以是内燃机、电动机、空气压缩引擎或其他类型的引擎组合，例如汽油发动机和电动机组成的混动引擎，内燃引擎和空气压缩引擎组成的混动引擎。引擎141将能量源142转换成机械能量。
能量源142的示例包括汽油、柴油、其他基于石油的燃料、丙烷、其他基于压缩气体的燃料、乙醇、太阳能电池板、电池和其他电力来源。能量源142也可以为车辆100的其他系统提供能量。
传动装置143可以将来自引擎141的机械动力传送到车轮144。传动装置143可包括变速箱、差速器和驱动轴。在一个实施例中，传动装置143还可以包括其他器件，比如离合器。其中，驱动轴可包括可耦合到一个或多个车轮121的一个或多个轴。
车辆100的部分或所有功能受计算平台150控制。计算平台150可包括至少一个处理器151，处理器151可以执行存储在例如存储器152这样的非暂态计算机可读介质中的指令153。在一些实施例中，计算平台150还可以是采用分布式方式控制车辆100的个体组件或子系统的多个计算设备。
处理器151可以是任何常规的处理器，诸如商业可获得的CPU。替选地，处理器151还可以包括诸如图像处理器(Graphic Process Unit:GPU)，现场可编程门阵列(Field Programmable Gate Array:FPGA)、片上系统(Sysem on Chip:SOC)、专用集成芯片(Application Specific Integrated Circuit:ASIC)或它们的组合。尽管图1功能性地图示了处理器、存储器、和在相同块中的计算机110的其它元件，但是本领域的普通技术人员应该理解该处理器、计算机、或存储器实际上可以包括可 以或者可以不存储在相同的物理外壳内的多个处理器、计算机、或存储器。例如，存储器可以是硬盘驱动器或位于不同于计算机110的外壳内的其它存储介质。因此，对处理器或计算机的引用将被理解为包括对可以或者可以不并行操作的处理器或计算机或存储器的集合的引用。不同于使用单一的处理器来执行此处所描述的步骤，诸如转向组件和减速组件的一些组件每个都可以具有其自己的处理器，所述处理器只执行与特定于组件的功能相关的计算。
在此处所描述的各个方面中，处理器可以位于远离该车辆并且与该车辆进行无线通信。在其它方面中，此处所描述的过程中的一些在布置于车辆内的处理器上执行而其它则由远程处理器执行，包括采取执行单一操纵的必要步骤。
在一些实施例中，存储器152可包含指令153(例如，程序逻辑)，指令153可被处理器151执行来执行车辆100的各种功能。存储器152也可包含额外的指令，包括向信息娱乐系统110、感知系统120、决策控制系统130驱动系统140中的一个或多个发送数据、从其接收数据、与其交互和/或对其进行控制的指令。
除了指令153以外，存储器152还可存储数据，例如道路地图、路线信息，车辆的位置、方向、速度以及其它这样的车辆数据，以及其他信息。这种信息可在车辆100在自主、半自主和/或手动模式中操作期间被车辆100和计算平台150使用。
计算平台150可基于从各种子系统(例如，驱动系统140、感知系统120和决策控制系统130)接收的输入来控制车辆100的功能。例如，计算平台150可利用来自决策控制系统130的输入以便控制转向系统133来避免由感知系统120检测到的障碍物。在一些实施例中，计算平台150可操作来对车辆100及其子系统的许多方面提供控制。
可选地，上述这些组件中的一个或多个可与车辆100分开安装或关联。例如，存储器152可以部分或完全地与车辆100分开存在。上述组件可以按有线和/或无线方式来通信地耦合在一起。
可选地，上述组件只是一个示例，实际应用中，上述各个模块中的组件有可能根据实际需要增添或者删除，图1不应理解为对本申请实施例的限制。
在道路行进的自动驾驶汽车，如上面的车辆100，可以识别其周围环境内的物体以确定对当前速度的调整。所述物体可以是其它车辆、交通控制设备、或者其它类型的物体。在一些示例中，可以独立地考虑每个识别的物体，并且基于物体的各自的特性，诸如它的当前速度、加速度、与车辆的间距等，可以用来确定自动驾驶汽车所要调整的速度。
可选地，车辆100或者与车辆100相关联的感知和计算设备(例如计算系统131、计算平台150)可以基于所识别的物体的特性和周围环境的状态(例如，交通、雨、道路上的冰、等等)来预测所述识别的物体的行为。可选地，每一个所识别的物体都依赖于彼此的行为，因此还可以将所识别的所有物体全部一起考虑来预测单个识别的物体的行为。车辆100能够基于预测的所述识别的物体的行为来调整它的速度。换句话说，自动驾驶汽车能够基于所预测的物体的行为来确定车辆将需要调整到(例如，加速、减速、或者停止)什么稳定状态。在这个过程中，也可以考虑其它因素来确定车辆100的速度，诸如，车辆100在行驶的道路中的横向位置、道路的曲率、静态和动态物体的接近度等等。
除了提供调整自动驾驶汽车的速度的指令之外，计算设备还可以提供修改车辆100的转向角的指令，以使得自动驾驶汽车遵循给定的轨迹和/或维持与自动驾驶汽车附近的物体(例如，道路上的相邻车道中的轿车)的安全横向和纵向距离。
上述车辆100可以为轿车、卡车、摩托车、公共汽车、船、飞机、直升飞机、割草机、娱乐车、游乐场车辆、施工设备、电车、高尔夫球车、火车等，本申请实施例不做特别的限定。
图2示出了一种在一些实施例中自动驾驶系统的框架示意图，从逻辑功能上来说，自动驾驶系统主要包括感知模块、规划模块和控制模块。
自动驾驶系统又可以称为自动驾驶系统(automated driving system，ADS)或驾驶辅助系统，例如，高级驾驶辅助系统(advanced driving assistant system，ADAS)。自动驾驶系统利用车辆上的传感器获取车辆自身的信息和车辆周围的信息，并对获取的信息进行分析和处理，实现例如障碍物感知、目标识别、车辆定位、路径规划、驾驶员监控/提醒等功能，从而提升车辆驾驶的安全性、自动化程度和舒适度。
感知模块具有环境感知的能力，例如可以识别车辆周围的障碍物，检测道路标记，识别信号灯，对车辆周围的行人/车辆的行为进行检测等。为了提升自动驾驶系统对环境的感知能力，车辆上设置的传感器种类和/或数量越来越多，例如激光雷达、毫米波雷达、和摄像头等。感知模块还可以具有车辆自身感知的能力，对车辆自身状态的进行感知的传感器的种类更加繁多，包括用于测量速度、温度、压力、流量、位置、气体浓度、光亮度、干湿度、距离等功能的传感器。例如，感知模块可以通过速度传感器获得车辆的速度，通过位置传感器获得加速踏板或制动踏板的位置。此外，感知模块可以通过定位(localization)系统确定车辆相对于环境的位置。定位系统例如为全球定位系统(global positioning system，GPS)和/或惯性导航系统(inertial navigation system)。
感知模块通过传感器获取车辆自身的信息和车辆周围的信息，并可以将获取的信息进行处理后提供给规划模块，该处理例如包括多传感器的信息融合(information fusion)，以提高感知信息的准确性。此时，感知模块也可以称为感知融合模块。
规划模块，其基于感知模块提供的信息做出驾驶决策/规划。例如，根据车辆当前所处的环境信息，做出车辆下一步驾驶行为的决策，例如，加/减速、变道、转向、刹车、或警示等。此外，自动驾驶系统还可以通过人机交互实现和车上人员(包括驾驶员或乘客)之间的信息交互，获取车上人员的需求，向车上人员反馈车辆当前状态或告警等。规划模块还可以进行路径决策/规划，例如，基于用户的需求完成最优路径的选择，用户的需求包括例如起点、终点、途径地、路径偏好等中的一个或多个。
控制模块基于规划模块的驾驶决策做出相应的控制，例如向对应的执行器或执行器的控制器发送控制指令，以控制执行器执行相应的动作，例如加/减速、变道、转向、刹车、或警示等。
自动驾驶系统的输入不仅包括自身传感器的信息输入，还可以从他处获得信息输入，以提升自动驾驶的性能。例如，自动驾驶系统可以从地图供应商的服务端获取地图信息，以辅助驾驶决策；再如，自动驾驶系统可以从云端获得驾驶环境信息，例如天气信息，交通状态信息(例如车流量、车流平均速度等信息)，以辅助驾驶决策； 再如，自动驾驶系统可以从其他智能终端(例如其他智能车，行人的便携终端等)获取他车的行驶信息和/或感知信息，以辅助驾驶决策。自动驾驶系统可以通过无线通信技术从他处获取信息。该无线通信技术例如为基于蜂窝网的通信或专用短距离通讯(dedicated short range communications，DSRC)，基于蜂窝网的通信例如为长期演进(long term evolution，LTE)通信技术或第五代(5th generation，5G)通信技术。其中，基于蜂窝网的通信包括车与任何事物(vehicle-to-everything，V2X)通信，V2X通信包括车辆与车辆之间(vehicle to vehicle，V2V)的通信、车辆与路边基础设施之间(vehicle to infrastructure，V2I)的通信、车辆与行人之间(vehicle to pedestrian，V2P)的通信或者车辆与网络之间(vehicle to network，V2N)的通信等，为车辆提供了多种获取信息的途径，丰富了自动驾驶系统的输入信息，提升了自动驾驶的性能。
此外，自动驾驶系统的实现需要较大的算力，往往通过专用的硬件平台来实现，例如多域控制器(Multi Domain Controller，MDC)，即以上感知模块、规划模块和控制模块的全部或部分功能通过专用的硬件平台，例如MDC，来实现。而自动驾驶系统中软件可以进行升级，例如感知模块对信息处理的算法，或规划模块的决策算法等都可以通过软件更新的方式进行升级，该软件更新可以通过空中升级(over the air，OTA)技术实现。
应当理解的是，上述“智能驾驶车辆”一般指具有SAE分级在L2及以上的车辆。
在下文的叙述中，一般以“自车”代指包括自动驾驶系统的主体车辆，应当理解的是：自车是以驾驶者所处的车辆来指定的便于说明的代称。
参见图3-1，其示例性地示出了在十字路口下自车的左转场景，图3-1的右侧示出了自车在左转专用车道上等待左转，在自车周围示出了四个障碍物，分布为：A车、B车、行人P1和行人P2。自车的自动驾驶系统会对这四个障碍物进行分析并分别作出它们的预测的轨迹，如图1所示意分别为：A车的预测轨迹为自南向北直行通过路口，B车的预测轨迹为自西向东直行通过路口，行人P1的预测轨迹为自北向南通过斑马线，行人P2的预测轨迹为自东向西通过斑马线，而自车执行左转的“预测的运动轨迹”如图示。虽然大部分路口都有交通指示灯，如果所有的交通参与者都严格按照交通灯的指示运动，从理论上而言所有的交通参与者之间是可以没有冲突的。但是在某些情况下：例如：1.在一些没有安装交通指示灯的路口；2.交通指示灯出现了故障，3.夜晚23点之后，交通指示灯可能会转换成仅为黄灯闪烁示意注意慢行的情况。在上述这些情况下的时候，自车与各个交通参与者(对自车而言为障碍物)之间可能会产生较多的冲突(即如图3所示意的C1-C4)。而为了避免发生冲突以引发交通事故，自车需要实时、不间断地跟踪这些交通参与者，并依据各个交通参与者的实时情况来实时地调整自动驾驶策略(是抢行，让行还是制动刹车)。这种情况会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
还应理解的是：在车辆进行左转U型掉头的情况下，由于车辆的可行驶空间范围往往较小(U型掉头需要较大的转弯半径和空间)，并且车辆U型掉头的轨迹往往会 和斑马线上的行人轨迹产生两次可能的冲突(交叉)，因此上述各种情况会愈发严重。参见图3-2所示意，其示出了自车在进行U型掉头的场景，图3-2中示出了行人P1以及自车的自动驾驶系统对行人P1的运动轨迹的预测；应当理解的是：虽然图3-2中仅示出了一个行人P1，但是如上所述，在实际的交通场景中，可能会存在数量众多(例如数十个甚至几十个行人穿越斑马线的情况)；可以看出：自车的预测的掉头轨迹和行人P1的预测轨迹之间有两处可能的冲突位置C1和C2。图3-2中还示出了自车的不可行驶区域A，自车的不可行驶区域A是由于车辆自身的尺寸以及车辆转向半径所确定的、在自车进行U型掉头的时候所无法进入的区域，也即是说：在自车进行U型掉头的时候，道路上留给自车可以进行规划行驶路径的区域是较少的。综合这些情况，可以看出：当自车在进行U型掉头的时候，不仅同样会产生上述问题，并且上述问题的严重程度会更加剧烈。
应当理解的是：图3-1和3-2仅是一些示例性的场景示意，在实际的城区交通场景中，障碍物的数量一般为十几个至几十个，而在城区路口的场景时，甚至可以达到数百个之多。图4示出了某自动驾驶车辆在实际场景下一段时间内所跟踪的目标(障碍物)数目，可以看出：峰值时的目标数量几乎可以达到一百，这种情况会对车辆自动驾驶系统的算力需求产生极大的峰值冲击，并导致上述的各种问题。
还应理解的是：即使是在路口配备了交通指示灯并且交通指示灯正常运作的情况下(即交通指示灯依据合理次序显示红、绿、黄)。仍然存在有交通参与者不严格按照交通指示灯的指示而运动的情况，典型的场景包括：车辆闯红灯，车辆闯黄闪(即黄灯闪烁、即将变成红灯的情况下车辆加速通过路口)，行人闯红灯等。因此即使在路口的交通指示灯正常运作的情况下，在路口处车辆的自动驾驶系统仍然要保持对各个交通参与者的持续监控和预测，因此车辆的自动驾驶系统的算力需求仍然处于较高位置，并且也会产生上述的各种问题。
图5示出了在环岛时的车辆驶出环岛的场景，参见图5，自车计划在下一个环岛路口按照预测轨迹驶出环岛，在自车左前方有A车，在自车右后方有C车，在另一个路口有B车即将驶入环岛。从图5可以看出，一方面，自车的预测轨迹和B车的预测轨迹和C车的预测轨迹都有冲突的可能，另一方面，自车的预测轨迹还和左前方的A车有冲突的可能。本领域人员应当理解，图5仅是为了示例而给出了自车和周围的A,B,C车辆，在实际的环岛场景中，自车周围的车辆数目可能为数十辆甚至几十辆，并且，由于环岛场景下，各车辆的驶出、驶出环岛的出/入口并不一致，更容易导致发生冲突，因此，环岛场景也会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
即使在一些看起来较为“简单”的场景结构中，仍然可能会出现和上述所类似的问题，如图6所示，在一个Y形路口，自车希望可以经过路口区域左转至Y形路口的一个分支，在自车的对向的A车预计经过路口区域继续直行，在Y形路口的一个分支处B车预计经过路口区域驶入自车的当前车道的对向车道，在自车的右前方还有行人P1计划穿过人行道。从图6可以看出，自车的左转预测轨迹和上述的A车、B车以及 行人均由冲突的可能。本领域人员应当理解，图6仅是为了示例而给出了自车和周围的A,B车辆和行人P1，在实际的场景中，自车周围的车辆数目可能为数十辆甚至几十辆，而行人数目也有可能是数十名甚至几十名。因此，即使在这种看似“简单”的Y形路场景也会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
在如图7所示意的丁字路口场景中，自车预计经过路口区域左转，A车,B车,行人P1和P2的预测轨迹分别如图7所示意,从图7可以看出，自车的左转预测轨迹和上述的A车,B车以及行人P1,P2均有冲突的可能。本领域人员应当理解，图7仅是为了示例而给出了自车和周围的A,B车辆和行人P1,P2，在实际的场景中，自车周围的车辆数目可能为数十辆甚至几十辆，而行人数目也有可能是数十名甚至几十名。因此，即使在这种看似“简单”的丁字路口场景也会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
即使在高速/高架等较为简单的场景下，仍然可能存在自动驾驶系统无法实时、及时地处理当前信息的情况，参见图8，其示出了自车沿匝道汇入高速的场景示意图。在当前时刻，自车沿着匝道将要汇入高速车道，在自车前方的匝道上有同样意图的A车，在高速车道上有B车，C车，D车和E车；自车的自动驾驶系统在当前时刻需要完成的任务包括：判断匝道即将终止的区域(即图8中的道路终止域)，判断自车的预测轨迹和A车是否有冲突，判断自车的预测轨迹和高速车道上的B车，C车，D车和E车是否有冲突，在图8所示意的场景中，B车和C车沿着当前所在车道继续行驶，而D车因为前向E车速度较慢，因此计划切换车道，即汇入B车和C车所在的车道。而D车计划汇入B车和C车所在车道的行为会对B车和C车造成影响，即影响到B车和C车的预测轨迹，而B车和C车的预测轨迹的变化又进一步会导致A车和自车的预测轨迹，从而使得自车的自动驾驶系统需要不断地重新规划未来的行驶策略和行驶路径，从而使得自动驾驶系统对算力的需求保持在较高水平。应当理解的是，虽然图8示出了自车之外的A车-E车；但是这仅是一种示例性的示意，在节假日高峰时段时(例如国庆节假期)，高速道路上往往会有大规模车流，这些车流的车辆之间的相互影响将会使得上述的场景更加复杂，进一步地，这种场景会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
应当理解的是：在上述示例中，的自车可以通过感知系统来确定车道终止域，感知系统可以包括摄像装置和激光雷达，感知系统基于摄像装置获取的图像信息和激光雷达获取的点云信息进行传感器信息融合，从而判断车道终止域，并基于车道终止域和自车之间的距离、方位提前对自车进行轨迹规划。还应当理解的是，自车对于上述D车的转向意图的获取既可以基于D车的转向灯信息获取，也可以基于V2X(Vehicle to Everything)/V2V(Vehicle to vehicle)技术获取，还可以由自车的自动驾驶系统对 于D车在当前时刻之前的一段时间的行驶信息来确定(例如：如果在自车的感知系统的感知范围内的历史运动系统中有过超车行为并一直保持较高的运动速度，而且当前时刻D车的前方车辆保持较低的运动速度，则自车的自动驾驶系统可以认为D车有较强的超车意愿，并基于这种较强的超车意愿来预测D车的未来预测轨迹)，或者基于上述的多种方式结合的方式来确定。
参见图9，其示出了自车将要从高速车道下匝道的场景示意图。自车计划从右侧第二车道向右变道至最右侧车道并下匝道；在最右侧车道有A车和C车沿当前车道行驶，在自车所处车道前方有B车计划向右变道至最右侧车道并下匝道。由于B车处于A车和C车行进方向的前方，因此B车的变道下匝道行为会对A车和C车产生影响，并进而进一步影响到自车的未来行驶轨迹。应当理解的是，虽然图9示出了自车之外的A车-C车；但是这仅是一种示例性的示意，在节假日高峰时段时(例如国庆节假期)，高速道路上往往会有大规模车流，这些车流的车辆之间的相互影响将会使得上述的场景更加复杂，进一步地，这种场景会对自车的自动驾驶系统提出实时的高算力需求，这种高算力需求而这往往会导致自车的自动驾驶系统无法正常运作而产生卡顿，从而导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。
通过上述的各种场景示例可以看出：在自动驾驶领域，当自车的自动驾驶系统在一些“复杂困难”的场景时，容易因为周围目标(障碍物)的数量和种类众多，实时的对目标的跟踪需求高，以及目标运动状态的随时发生变化等综合因素所产生的对自动驾驶系统算力的高需求冲击，这种高需求会冲击导致自车的自动驾驶系统不能实时、及时地处理当前的情况，无法规划出实时有效的自动驾驶策略。这些场景包括但不限于十字路口左转、丁字路口左转、环岛驶出、Y形路口左转、通过匝道上高速/高架、高速/高架下匝道等等。在这些场景下，如果路口处的交通指示灯处于故障状态或者退化的状态(例如在夜晚23点之后，交通指示灯可能退化为一直保持黄闪状态)；或者当有些交通参与者并不遵守交通规则的情况下，上述的问题会变得更加严重。
对于上述问题，本申请提供一种综合的解决方案，本申请的主要创新包括：(1)将自动驾驶系统的任务按照不同的类型固定匹配给自动驾驶系统的处理器核(2)对车辆周围的障碍物采用交互博弈的处理策略
下面详细介绍本申请的实施方式。
车辆的自动驾驶系统在运行时，一般包括不同的任务类型：例如：操作系统任务、自动驾驶任务、数据落盘任务、数据上云任务、资源调度任务等等；其中，参见图2和相关的说明，自动驾驶任务可以分为感知任务、决策/规划任务和控制任务。
从自动驾驶系统的提供算力的硬件而言，目前各自动驾驶系统一般都包括多个处理器，每个处理器可能包括多个处理器核。图10示出了一种自动驾驶系统的处理器核的示意图，该自动驾驶系统包括了双SOC芯片A和B，每个芯片A和B都包括16个核，因此，该自动驾驶系统共包括32个处理器核。
在一些实施例中，将自动驾驶系统的任务按照类型固定分配/匹配给上述的32个处理器，可以按照如下的优先级来进行分配，(1)操作系统任务，(2)自动驾驶任务，(3)数据落盘(存储)任务和数据上云任务，(4)资源调度任务和资源配置任 务等。更具体地：例如可以将S0C A和B的0-1号处理器核分配给操作系统任务；将SOC A和B的2-13号处理器核分配给自动驾驶任务；将SOC A和B的14号处理器核分配给数据落盘和数据上云任务；将SOC A和B的15号处理器核分配给资源调度任务和资源配置任务。
本领域人员应当理解，上述的处理器核的数目和固定分配的模式仅是一种示例，本领域人员可以在合理的范围内改变上述处理器的核的数目和固定分配的模式。例如自动驾驶系统可能包括三个处理器，每个包括16个处理器核；又例如可以将资源调度任务和资源配置任务固定配置三个处理器核。
上述的“自动驾驶系统的任务固定分配给自动驾驶系统的处理器核”指的是：以上述的自动驾驶任务而言，一共分配了24个处理器核(SOC A和B的2-13号处理器核)，则车辆在行驶过程中时，其自动驾驶系统的自动驾驶任务(感知、规控和控制)均由这24个处理器核进行处理；自动驾驶任务不会由其它的处理器核进行处理，另外，其它的非自动驾驶的任务(例如操作系统、资源调度等)也不会由这24个处理器核进行处理。
对于其它种类的任务和处理器核之间的“固定分配”，也是如此。即不同种类的任务固定由设定好的处理器核进行处理。
可以看出，在上述的设置中，自动驾驶任务被分配了最多数量的处理器核(24个)；通过这种设置,可以保证在某些场景下，当目标(障碍物)的数目众多的时候，对自动驾驶系统的高算力需求可以得到满足。本领域人员应当理解：在目标(障碍物)的数目众多的时候，对自动驾驶系统的任务的需求主要来自对目标障碍物的实时动态的跟踪，以及基于对目标障碍物的实时动态跟踪所规划出的自动驾驶策略，即这种时候的主要任务在自动驾驶任务中的“感知”和“规划”部分。而通过预先将自动驾驶系统的处理器核中的大多数固定分配给自动驾驶任务，可以保证即使在上述的多目标(障碍物)的场景下，上述24个核所提供的算力足够可以满足在上述各种场景下的自动驾驶任务中的感知和规划任务。从而避免了因为不能即时的处理感知和规划任务所导致的无法即时地给出自动驾驶策略的问题。
在上述的设置中，对于非自动驾驶任务固定分配了较少的处理器，这些任务包括操作系统和数据落盘任务等，应当理解的是，为这些任务固定分配的处理器核的数目虽然少于固定分配给自动驾驶任务的处理器核的数目，但是仍然足够处理上述任务。还应当理解的是，在上述的对算力高需求的各种场景下，对这些非自动驾驶任务同样也是高需求的，例如当自车周围的目标(障碍物)数目较多的时候，所需存储的数据量也随之增加。但是，通过对这些非自动驾驶任务固定分配处理器核，使得这些非自动驾驶任务的算力需求不会延伸出固定分配的处理器核，从而影响自动驾驶任务的算力需求。
在一些实施例中，在自动驾驶任务中，除了上述的感知任务和规划任务，还包括例如GNSS(Global Navigation Satellite System)定位任务，地图(或者高精度地图)服务任务等。在上述的各种对算力需求高度高的场景下，相对于感知任务和规划任务，定位服务和地图服务的实时性需求较低，因此可以对自动驾驶任务进行优先级排序，将实时性需求高的感知和规划任务的优先级设定为高，将定位和地图服务的优 先级设定为中等，在对上述自动驾驶任务进行算力分配的时候，优先满足优先级较高的任务，这样，可以保证在对算力需求高的场景下，对于自动驾驶任务，固定分配的处理器核优先处理“紧急”的任务(例如感知任务和规划任务)，以保证自动驾驶任务的顺利进行和车辆的安全驾驶。
在一些实施例中，在自动驾驶任务中，可以对感知任务和规划任务以其所具体实现的功能为基准进行进一步的拆分。以感知任务而言，可以包括例如LD(Lane Detection:车道线检测)，PD(Pedestrian Detection:行人检测)，VD(Vehicle Detection:车辆检测)，RL(Radar and LiDar感知融合)等；以规划任务而言，可以包括例如Router(全局路径规划)，FSM(Finite Status Machine:有限状态机)，MOP(Motion Planning:运动控制)等。对于上述功能，可以进一步地固定分配处理器核；例如，可以对LD功能固定分配上述24个处理器核中的两个，对MOP功能固定分配上述24个处理器核中的一个。应当理解的是，如果上述的具体实现的功能的数目过多以至于无法满足对于每个具体实现的功能都至少被固定分配给一个处理器核。则可以将上述所具体实现的功能按照优先级进行排序，对于高优先级的功能，固定分配所对应的处理器核，对于低优先级的任务，可以不固定分配处理器核。
应当理解的是，上述的自动驾驶任务中的“高优先级”或“低优先级”的任务可以是预先设定好的，也可以是被依需要而人工设定的，还可以是依据车辆所处的环境而自动进行调整的。还应当理解的是，上述的24个固定分配给自动驾驶任务的处理器核中的处理器核的数目仅仅是示例，实际中也可以是其它的数目，例如36个、48个或者其它合适的自然数。
例如，仍以固定分配24个处理器核给自动驾驶任务为例，可以在自动驾驶系统装车前进行预先设定，固定分配给VD(车辆检测)两个处理器核，而不对PD任务固定分配处理器核。
又例如，可以在自动驾驶系统中设有用户设置界面，用户可以自行设定一至两个固定分配处理器核的任务，例如，如果用户认为RL任务比较重要，可以固定分配一个处理器核给RL任务。
又例如：在不同的车辆运行区域，同一个任务的重要程度也会不同，例如在高速路场景而PD(行人检测)并不重要(因为高速路段上一般而言不会有行人)；而在城区场景下，因为行人众多，PD任务就是一个基础而重要的任务。因此，在高速场景下，可以不对PD任务固定分配处理器核；在城区场景下，对PD任务固定分配处理器核。
上述的各种方式(预先设定，人工设定，依据车辆所处场景自动调整)也可混合使用。例如，可以在自动驾驶系统装车前进行预先设定，并保留依据车辆所处场景进行自动调整的功能。
参见图11，其示例性地给出了一个十字路口场景下的情况，自车意图左转停在左转待行区，在此场景中，自车的感知系统需要进行感知的目标(障碍物)包括：
车道线；
停止线；
导向箭头；
交通信号灯；
交通指示牌；
斑马线；
自车周围的车辆(A车，B车，C车，D车，E车，F车，G车；上述A-G车仅为示例，在实际场景中自车周围的车辆数目可能为数十辆甚至几十辆)；当自车周围有较大的车辆的时候(图8中未示出)，较大的车辆还有可能对自车造成感知盲区，例如遮蔽了自车的摄像装置获取交通信号灯的信号；
行人P1(图8中的行人P1仅为示例，在实际场景中自车周围的行人数目可能为数十人甚至几十人)；
道路上的路障/施工标识；
道路周围的静止障碍物(例如树木)；
道路周围的路侧设备RSU(Road-Side Unit)；
道路上的静止障碍物(例如道路上的碎石、坑洼等)；
两轮机动车；
可以看出，在图11所示例的场景中，自动驾驶系统需要感知的目标不仅数量众多，而且种类繁多，因此在图10所示例的场景以及相类似的场景下，自动驾驶系统的自动驾驶任务中的感知任务对算力的需求将达到较高的水平。
进一步地，自车的规划任务将需要基于感知系统的信息进行自动驾驶策略规划，策略规划中需要考虑的因素包括：
交通信号灯和交通指示牌的信号指示；
自车和周围其它车辆的预测轨迹的可能的冲突；
自车是否处于车道线和停止线所规定的范围内；
自车需要规避道路上的静止障碍物以及路障/施工标识；
自车所接收到的路侧设备的信息；
自车和行人的预测轨迹的可能的冲突；
自车和道路上的其它机动车辆(例如两轮机动车)的预测轨迹的可能的冲突；
可以看出，在图11所示例的场景中，自动驾驶系统的规划系统也需要处理繁多的信息才能做出相应的自动驾驶策略。例如，自车的规划系统在规划路径的时候需要规避静止障碍物和路障/施工标识；需要避免和周围行人以及两轮激动车发生碰撞；需要避免和周围的其它车辆A-G的轨迹发生冲突，等等。尤其是，在实际的城区工况中，自车周围的车辆数目和行人数目可能达到几十之多，自车需要处理的自动驾驶任务的算力需求因此会达到高峰状态。
而在一些情况下，如果图11中的交通信号灯失效或者降级，将导致上述的自动驾驶任务的算力需求进一步加剧。
在一些实施例中，采用上述的固定分配处理器核的策略，仍以上述的分配24个处理器核给自动驾驶任务为例，可以将对VD和PD任务进一步分配处理器核，以保证自车的自动驾驶系统对于行人周围车辆检测和行人检测任务的算力需求可以得到满足；另外，还可以对Router和MOP任务固定分配处理器核，以保证自车的路径规划和运动控制任务的算力需求可以得到满足。通过这种设置，可以保证在此类对算力需求达到高峰的场景时，自动驾驶任务可以即时顺畅地被完成。
在一些实施例中，在进行了上述固定分配处理器核的策略后，可以进行自车周围的关键目标进行选取，并且自动驾驶系统将关键目标的感知和相关规划设定为较高优先级的任务，关键目标选取的策略可以依据道路交通参与者的强势与弱势来确定，例如，如果车辆和行人发生碰撞可能会导致较严重的人伤后果，因此，可以将例如图11中的行人P1作为关键目标，自动驾驶系统以较高优先级对行人P1进行跟踪和轨迹预测，并做出相应的自车轨迹规划。
在一些实施例中，考虑到交通参与者的强势和弱势情况，可以进一步地将道路上的非机动车(例如自行车)作为关键目标，自动驾驶系统以较高优先级对非机动车进行跟踪和轨迹预测，并做出相应的自车轨迹规划。
在一些实施例中，对关键目标的选取可以包括交通参与者中较容易发生事故方，例如，在目前的社会经济生活中，快递是一种常见的工作类型，在城市中，从事餐饮快递工作的人数较多，而从事餐饮快递的人一般采用两轮(机动或者电动)车辆作为送餐工具，并且易于发生交通事故。因此，在关键目标的选取上，可以将道路上的两轮车作为关键目标，例如图11中的两轮机动车，自动驾驶系统以较高优先级对两轮机动车B1进行跟踪和轨迹预测，并做出相应的自车轨迹规划，以避免发生交通事故。
在一些实施例中，考虑并不是道路上所有的两轮车都是由送餐员所驾驶的，并且，目前的送餐员一般会穿着颜色鲜明的制服(如蓝色、黄色等)，因此可以对自动驾驶系统的感知系统进行预先训练，具体地，如果感知系统使用了神经网络(例如卷积神经网络)来处理所获取的图像信息，则预先对神经网络使用包括送餐员穿着制服的图像进行训练，当训练完成后(即感知系统的神经网络对送餐员的识别准确率超过预设值)，即可在车辆上使用感知系统识别送餐员，并将送餐员作为关键目标，以避免和他们发生交通事故。在一些实施例中，例如可以采用ResNet-50作为感知系统所使用的神经网络，应当理解的是，也可以采用其它类型的神经网络(例如VGG)作为感知系统所使用的神经网络,只要在符合本申请所披露的精神内使用即可。
在一些实施例中，进一步地考虑到送餐员往往可能会有创红灯或者逆行的情况发生。因此，在自动驾驶系统将送餐员作为关键目标之后，实时地对送餐员进行不间断的目标跟踪和轨迹预测，自动驾驶系统并规划出避免和送餐员发生碰撞的自车轨迹。
在一些实施例中，在自车确定了各关键目标后，还可以在中控屏上对关键目标进行突出/高亮/着重显示，以提醒自车驾驶员注意上述自车周围的关键目标。
在一些实施例中，在自车和各关键目标之间的碰撞风险超过设定阈值时，提醒驾驶员接管自车。可以采用声学、光学或者结合的方式来对驾驶员进行提醒。
通过对道路交通参与者中选取关键目标，以及通过对自动驾驶系统的感知系统进行预先训练。可以使得自车在上述的各种复杂的交通场景下，尽可能地避免与VRU(Vulnerable Road Users:弱势道路使用者)发生碰撞以造成人身伤害，另一方面，自车还可以尽可能地避免与容易发生交通事故的道路交通参与者(例如送餐员)发生碰撞。从而使得自车在复杂的城区道路环境中具有较高的自动驾驶安全性。
应当理解的是：上述关键目标的选取和相应的自车路径规划是建立在本申请所揭示的固定分配处理器核给自动驾驶任务的前提下的。在真实的场景下，道路上的行人和非机动车的数目可能有几十甚至上百之多，可以将关键目标的感知和与它们对应的 规划任务赋予较高的任务优先级，在一些实施例中，也可以将对关键目标进行感知和规划的任务固定分配处理器核，从而最大可能地保证车辆在上述复杂场景中的自动驾驶安全性。
在一些实施例中，自车的感知系统的能力可能会被周围的大型障碍物所影响，例如，自车的摄像装置可能会被自车前方的大型车辆(例如大巴车、货车等)所遮蔽，从而无法获得前方路口的交通信号灯的信号。在这种情况下，自车可以通过和路侧设备的交互，基于路侧设备和交通信号灯之间的信息交互，实时地获得前方交通信号灯的信号。
在一些实施例中，当自车处于各种复杂困难场景时，自动驾驶系统会对该场景下的交通信号灯进行检测；如果自动驾驶系统确定当前场景下的交通信号灯处于不工作或者降级工作的状态(例如前述的仅仅显示黄闪)；则自动驾驶系统对自车进行限速，例如将车速限制在40km/h以下。当交通信号灯不工作或者降级工作的状态时，在上述各种复杂场景下，交通信号灯所给出的信息量较少甚至为零，从而难以形成对所有的交通参与者的完备的指示，因此极易引发交通事故。而通过上述的在交通信号灯不工作或者降级工作时候的自车进行限速，从而使得自车在此类场景下以较低速度进行行驶，降低发生交通事故的可能。
应当理解的是:上述的自动驾驶系统确定当前场景下的交通信号灯处于不同做或者降级工作的状态可以由多种方式达成；例如，自车可以通过和路侧设备进行通信，路侧设备与交通信号灯保持通信，路侧设备将交通信号灯的状态发送给自车。或者，自动驾驶系统的感知系统可以包括神经网络(例如卷积神经网络)，并预先对神经网络进行训练，使得神经网络可以识别交通信号灯不工作或者降级工作的状态；并依据训练好的神经网络来判断交通信号灯是否工作。
上述的对自车进行限速仍然是建立在对自动驾驶任务固定分配处理器核的基础上的。进一步考虑，对自车进行限速也进一步利于自动驾驶系统有较为充足的时间对所获取的信息进行处理并作出规划，因为较低的车速意味着通过同样的距离需要更多的时间，通过对自车进行限速，可以使得自动驾驶系统在上述困难的场景下(交通路况并且红绿灯失效/降级)有更多的时间可以处理信息，做出决策。
在一些实施例中，进一步地考虑到自车周围的障碍物，自动驾驶系统在对周围的＝障碍物进行感知、跟踪和做出对应规划的时候，采用交互博弈的策略。具体地：首先要准确识别自车周围障碍物的类别、运动状态信息，并根据障碍物的历史运动状态以及道路拓扑信息，理解障碍物的意图，进而预测障碍物要达到其意图目标可能的行为轨迹，障碍物交互决策根据障碍物的不同意图下可能的行为轨迹以及自车的行为轨迹空间评估障碍物和自车可能的交互行为轨迹空间，进行交互决策。
应当理解的是，上述自动驾驶系统采用交互博弈的策略也是建立在本申请的对自动驾驶任务固定分配处理器核的基础上的，通过该设置，使得自车在处于上述各种复杂场景且周围动态障碍物众多的情况下，自动驾驶系统有足够的算力可以及时地处理对周围动态障碍物的感知任务和规划任务。
下面介绍自车和周围动态障碍物的交互博弈过程。
首先，自动驾驶系统判断自车周围的动态障碍物中属于博弈目标的部分和属于非博弈目标的部分。如果自车的未来运动轨迹和动态障碍物的未来运动轨迹有冲突，则将动态障碍物确定为博弈目标；反之将其确定为非博弈目标。
参见图12，假设自车的运动轨迹为直行，自车的自动驾驶系统根据获取到的感知信息确定当前范围内有2个目标(或称障碍物)，分别是A车和B车，且A车与B车处于运动状态。其中，自动驾驶系统可以根据道路拓扑和目标的运动状态生成A车的预测轨迹以及B车的预测轨迹。其中，B车的预测轨迹与自车的运动轨迹有冲突，而A车处于与自车同向直行的状态，A车的预测轨迹与自车的运动轨迹无冲突，因此，自动驾驶系统将B车确定为博弈目标，即自车在当前时刻的后续时段内的运动过程中会与B车进行博弈，自车后续的运动决策需要考虑B车的预测轨迹进行制定。
自动驾驶系统的感知模块在自车与博弈目标进行博弈的过程中，对博弈目标的运动状态，包括横向状态特征和纵向状态特征，进行持续的观测并记录关键信息，感知模块将这些信息发送给规划模块，规划模块依据所获取的信息生成决策方案集。
在一些实施例中，参见图13，规划模块可以根据获取到的感知信息，并考虑道路限速、加速度改变Jerk值、博弈目标类型等纵向特征信息，生成纵向加速度采样维度，考虑道路边界、静止障碍物、车辆运动学等特征信息，生成横向偏移采样维度；规划模块将横向位置偏移采样维度与纵向加速度采样维度组合，张成采样空间，张成的采样空间是所有策略的可能性空间。规划模块可以根据张成的采样空间得到所有决策方案的集合。
规划模块可以基于决策方案集中每个决策方案包括的自车与博弈目标动作对的类别，将决策方案集划分为至少一个决策方案子集。即：规划模块可以为决策方案集中的自车与博弈目标所有动作对标注决策标签，并将标签一致的一簇划分为一个决策方案子集，每个决策方案子集中包括至少一个决策方案。
参见图14，在一些实施例中，规划模块为决策方案集中的每个决策方案标注决策标签，例如，在如图14所示的实施例中场景中，对决策方案集中的每个决策方案标注决策标签后，决策方案集中的决策标签可以包括例如：让行无避让，抢行无避让，抢行避让以及让行避让。上述的各标签分别代表自车对待博弈目标的决策类型，例如：对于“让行无避让”决策，代表自车在后续的博弈过程中，通过减速来进行让行，但是并不显著地改变自己的运动方向(无避让)；又例如：对于“抢行避让”决策，代表自车在后续的博弈过程中，通过不减速或者加速来抢行，但是在抢行的过程中，较为显著地改变自身的行驶方向以避让博弈目标。
在一些实施例中，可以对上述的各决策标签所代表的决策类型进行定量评估，具体低，本申请实施例提供的一种用于计算决策方案对应收益的收益评价体系，可以包括决策方案的安全性、舒适性、通过性、路权、偏移以及历史决策结果等几个决策维度。
在一些实施例中，该收益评价体系中的决策收益可以根据决策代价值(cost)确定，cost越小，决策收益越高，则该决策方案作为最优决策方案的可能性越高，决策代价值的具体评价维度与解释如下：
1)安全性cost：本申请实施例中的安全性cost主要是基于自车运动轨迹推演与博弈目标运动轨迹推演过程中，自车与博弈目标之间的距离得到的一个决策代价值。
其中，自车与博弈目标之间的距离越小，安全性越低，也就是说自车与博弈目标之间 的距离越小，自车安全性cost就越大，则自车基于安全性的决策收益越小。
进一步的，本申请实施例中还可以设置一个自车与博弈目标之间相距的最小距离，当自车与博弈目标之间相距的距离逐渐接近该最小距离的过程中，该安全性cost逐渐增大，基于安全性的决策收益逐渐减小；当自车与博弈目标之间相距的距离不大于该最小距离时，该安全性cost达到最大，基于安全性的决策收益达到最小。
此外，本申请实施例中还可以设置一个自车与博弈目标之间用于计算安全性cost的最大距离，当自车与博弈目标之间相距的距离不小于该最大距离时，该安全性cost达到最小，基于安全性的决策收益达到最大。
在一些实施例中，如图15中的(a)所示，横轴表示自车与博弈目标之间的距离，横轴中的A点表示本申请实施例中设置的自车与博弈目标之间的最小距离，横轴中的B点可以表示本申请实施例中设置的自车与博弈目标之间用于计算安全性cost的最大距离，纵轴表示安全性cost。
其中，当自车与博弈目标之间相距距离不小于B时，安全性cost达到最小，基于安全性的决策收益达到最大；当自车与博弈目标之间相距距离在A点与B点之间时，自车与博弈目标之间相距距离越趋近A点，安全性cost越大，基于安全性的决策收益越小，自车与博弈目标之间相距距离越趋近B点，安全性cost越小，基于安全性的决策收益越大；当自车与博弈目标之间相距距离不大于A时，安全性cost达到最大，基于安全性的决策收益达到最小。
其中，本申请实施例上述图15中的(a)所示的A点以及B点可以是基于实际经验得到的。例如，本申请实施例上述图15中的(a)所示的A点可以为0.2m～0.5m(比如，A点为0.5m)，B点可以为1.0m～3.0m(比如，B点为2.5m)。
2)舒适性cost：本申请实施例中的舒适性cost主要是基于自车当前加速度与期望加速度的差值得到的一个决策代价值。
其中，自车当前加速度与期望加速度的差值越小，舒适性cost越小，则自车基于舒适性的决策收益越大。
进一步的，本申请实施例中还可以设置一个自车当前加速度与期望加速度的最大差值，当自车当前加速度与期望加速度的差值越接近该最大差值时，该舒适性cost越大，基于舒适性的决策收益越小。
在一些实施例中，如图9-4中的(b)所示，横轴表示自车当前加速度与期望加速度的差值，例如，自车当前加速度为1m/S^2，期望加速度为4m/S^2，则自车当前加速度与期望加速度的差值为3m/S^2。横轴中的A点表示本申请实施例中设置的自车当前加速度与期望加速度的最大差值。
其中，当自车当前加速度与期望加速度的差值越趋近A点时，舒适性cost越大，基于舒适性的决策收益越小；当自车当前加速度与期望加速度的差值越远离A点时，舒适性cost越小，基于舒适性的决策收益越大；当自车当前加速度与期望加速度的差值不小于A点对应的加速度值时，舒适性cost达到最大，基于舒适性的决策收益达到最小。
其中，本申请实施例上述图15中的(b)所示的A点可以是基于实际经验得到的。
例如，本申请实施例上述图15中的(b)所示的A点可以为5m/S^2～7m/S^2(比如，A点为6m/S^2)。
3)通过性cost：本申请实施例中的通过性cost主要是基于自车当前速度与自车到达冲突点(例如，自车与博弈目标轨迹可能发生交互的点)的速度之间的速度差值得到的一个决策代价值。
其中，速度差值(当前速度减去冲突点速度)越大，通过性cost越大，则自车基于通过性的决策收益越小。
进一步的，本申请实施例中还可以设置一个速度差值的最大值，当速度差值越接近该最大值时，该通过性cost越大，基于通过性的决策收益越小。
在一些实施例中，如图15中的(c)所示，横轴表示该速度差值，横轴中的A点表示本申请实施例中设置的速度差值的最大值。其中，当该速度差值越趋近A点时，通过性cost越大，基于通过性的决策收益越小；当该速度差值越远离A点时，通过性cost越小，基于通过性的决策收益越大；当该速度差值不小于A点对应的速度差值时，通过性cost达到最大，基于通过性的决策收益达到最小。
其中，本申请实施例上述图15中的(c)所示的A点可以是基于实际经验得到的。
例如，本申请实施例上述图15中的(c)所示的A点可以为6m/s～10m/s(比如，A点为7m/s)。
4)路权cost：本申请实施例中的路权cost是基于自车或者博弈目标的路权情况，针对自车当前的加速度得到的一个决策代价值。
其中，本申请实施例中主要基于高路权一方来进行计算，对有路权的那一方进行路权cost惩罚，例如，如果自车是高路权，那么如果自车减速的话，则进行路权cost惩罚。
可以理解的，本申请实施例中的路权是自车与博弈目标的相对的关系，当自车的路权高于博弈目标的路权时，由于自车的路权高，则交互博弈过程中应使得高路权的车尽量不改变运动状态。因此，若自车的加速度下降，做减速行驶的话，则自车该运动状态的改变应该施以较高的路权cost惩罚，即当基于自车路权高于博弈目标的路权情况下，自车当前加速度越小，路权cost越大，则自车基于路权的决策收益越小。
进一步的，本申请实施例中还可以设置一个自车当前加速度的最小值，当自车当前加速度越接近该最小值时，该路权cost越大，基于路权的决策收益越小。
在一些实施例中，如图15中的(d)所示，ACC表示加速度，其中，自车高路权是一个限定前提，例如，将自车高路权作为前提输入，基于自车与博弈目标当前的路权关系，进一步确定路权cost。
其中，横轴表示自车当前加速度，横轴中的A点表示本申请实施例中设置的自车当前加速度的最小差值。其中，当自车当前加速度越趋近A点时，路权cost越大，基于路权的决策收益越小；当自车当前加速度越远离A点时，路权cost越小，基于路权的决策收益越大；当自车当前加速度不大于A点对应的加速度值时，路权cost达到最大，基于路权的决策收益达到最小。
其中，本申请实施例上述图15中的(d)所示的A点可以是基于实际经验得到的。
例如，本申请实施例上述图15中的(d)所示的A点可以为(-1)m/S^2～(-3)m/S^2(比如，A点为(-1)m/S^2)。
5)偏移cost：本申请实施例中的偏移cost主要是基于自车相对参考线的位置偏移，得到的一个决策代价值。其中，可以理解的，自车相对参考线的偏移距离越远，则偏移cost 值越大，自车基于偏移的决策收益越小。
可选的，本申请实施例中的参考线可以为车辆所在道路的中心线。
进一步的，如图16所示，本申请实施例中可以包括两个偏移区域，例如区域A与区域B，本申请实施例中可以将区域A称为软边界，区域B可以称为硬边界。软边界可以为车辆所在的行驶车道的区域，也就是说，自车相对参考线发生偏移，但自车一直在该行驶车道中行驶，则可以理解为自车一直在软边界中；本申请实施例中设置的硬边界可以为超出行驶车道的区域，即自车相对参考线发生偏移，且自车驶离该行驶车道，则可以理解为自车进入到了硬边界中。
此外，由于车辆驶离该行驶车道时，车辆距离参考线的偏移更大，且行驶过程中的风险因素较多，安全较低，因此，本申请实施例中可以对车辆驶离该行驶车道，即车辆处于硬边界时的惩罚代价提高。例如，本申请实施例基于不同的区域可以采用不同的计算偏移cost的斜率值进行计算，其中，在软边界中设置的用于计算偏移cost的斜率值可以较小，在硬边界中设置的用于计算偏移cost的斜率值可以较大。
在一些实施例中，如图15中的(e)所示，横轴表示自车当前距离参考线的偏移距离，横轴中的-A点表示本申请实施例中设置的软边界的最左侧位置，0点表示参考线位置，A点表示本申请实施例中设置的软边界的最右侧位置；-B点表示本申请实施例中设置的硬边界的最左侧位置，B点表示本申请实施例中设置的硬边界的最右侧位置。其中，车辆在软边界中相对参考线的偏移cost采用斜率1进行计算，车辆在硬边界中相对参考线的偏移cost采用斜率2进行计算，且斜率2大于斜率1。
其中，当自车距离参考线0点的偏移距离越大时，偏移cost越大，基于偏移的决策收益越小。当自车距离参考线0点的偏移距离不小于C点对应的距离时，偏移cost达到最大，基于偏移的决策收益达到最小。
其中，本申请实施例上述图9-4中的(e)所示的A点，B点，以及斜率1与斜率2可以是基于实际经验得到的。
例如，本申请实施例上述图9-4中的(e)所示的A点可以为1.5m～2.5m(例如，A点为2m，则对应的-A点为-2m)，B点可以为1.9m～2.9m(例如，B点为2.9m，则对应的-B点为-2.9m)。
6)历史决策结果cost：本申请实施例中的历史决策结果cost也可以称为帧间关联代价cost，主要是基于自车上一帧的决策方案，得到的基于当前帧的一个决策代价值。
其中，可以理解的，对于同一博弈目标而言，假设自车上一帧是抢行，下一帧是让行的话，那么自车的速度改变情况较大，导致行驶波动比较大，舒适度降低。因此，为了有效的保障行驶波动较小，缓解波动，行驶的更平稳，之前如果是抢行的话，尽可能一直处于抢行状态。
在一些实施例中，如图15中的(f)所示，如果第K帧是抢行，那么第K+1帧还是抢行的话，历史决策结果cost较小，基于历史决策结果的决策收益较大，反之，如果第K帧是抢行，那么第K+1帧是让行的话，历史决策结果cost较大，基于历史决策结果的决策收益较小。
进一步的，本申请实施例中的该电子装置可以根据一定的权重比例，将上述各个评价维度的cost进行融合，可以得到基于多维度的决策代价值。
在一些实施例中，所述收益评价体系可以主要由历史决策结果确定，即可以主要依据历史决策结果(例如将历史决策结果cost的权重设置在80％-90％)，并根据历史决策结果确定车辆的驾驶风格，例如是保守、适中或者激进。
在一些实施例中，本申请实施例设置收益评价体系中各个评价维度对应的权重为：
安全性10000，通过性6000，舒适性500，路权1000，偏移2000，风险区域1000。
因此，本申请实施例得到的决策代价值可以为：
安全性cost*10000+通过性cost*6000+舒适性cost*500+路权cost*1000+偏移cost*2000+风险区域cost*1000的和。
进一步的，针对不同的行驶环境，收益评价体系中各个评价维度对应的权重可能并不相同。
在一些实施例中，行驶环境与评价维度权重的对应关系可以如下表1所示。
Figure PCTCN2022076331-appb-000001
表1：行驶环境与评价维度权重的对应关系
当车辆当前行驶环境处于市区道路时，策略可行域生成模块根据上述表1的内容，可以确定采用的收益评价体系中各个评价维度对应的权重为安全性cost*10000、通过性cost*6000、舒适性cost*500、路权cost*1000、偏移cost*2000、风险区域cost*1000。
通过上述cost评价体系，可以获取自车的决策收益值。
应当理解的是，上述的决策收益值可以按照归一化来处理，即理想的最优收益为1，各个策略的收益按照相互的比例进行归一化处理。
继续参见图14，在图9-3的各决策方案中，以虚线和实线示意出了不同的具体决策方案所对应的自车和博弈目标的轨迹。以“让行无避让”为例，在该决策方案下，自车和博弈目标各有三条对应轨迹，这三条轨迹对应于不同的决策收益值，其中自车的实线所标识的轨迹代表三条对应轨迹中自车决策收益值最大的那一个(也即是指：在目前让行无避让的策略评估中，自车按照该规划轨迹行驶是最“有利”的)，对应地，博弈目标的实线所标识的轨迹代表于自车的实线轨迹所对应的博弈目标的预测轨迹。将上述自车的实线轨迹称为“让行无避让”决策方案子集中的目标决策方案。相类似地，可以确定其它决策方案子集的目标决策方案。
应当理解的是：虽然图14中对于“让行无避让”示出了三个轨迹对(自车，博弈目标)，但是轨迹对的数目也可以是其它自然数，例如四个轨迹对，或者两个轨迹对等。
通过上述设置，可以确定每个决策方案子集中的目标决策方案。
在一些实施例中，通过对决策方案子集的目标决策方案进行进一步的评估，可以 将决策方案子集划分为不同的梯队，仍以图14为例，其中示出了四个决策方案子集：让行无避让，抢行无避让，抢行避让，让行避让。每个决策方案子集都有一个目标决策方案。通过对上述四个不同决策方案子集的目标决策方案进行决策收益值的比较，按照某一设定的阈值，可以将上述四个不同的决策方案子集划分为不同的“梯队”。例如，仍以上述四个目标决策方案为例，如果将阈值设定为0.76(阈值可以根据经验值或者历史值而获得)，如果抢行避让和让行无避让的决策收益值分别为0.8和0.77，它们均大于阈值；而让行无避让和让行避让的收益值分别为0.73和0.69，它们均小于设定的阈值。则可以将抢行避让和让行无避让的决策方案作为第一梯队，而将抢行无避让和让行避让作为第二梯队。回到物理世界，第一梯队代表着相对于第二梯队而言较优的“策略”。也即是说，在图14所示意的场景中(可以是自车直行，豁口有他车汇入的场景)，自车选择上述的抢行避让或者让行无避让的策略是较优的。从现实的物理世界角度去理解上述策略的话，对于上述的图14所示意的场景，如果自车选择抢行避让，则可以保证自车的良好的通行性，符合转弯让直行的基本交通准则，另外自车稍做横向避让以满足安全性的需求(以防止他车汇入的速度过快发生碰撞)，因此该策略既可以满足自车顺畅通行的需求，也保证了驾驶安全性，总体来说是一个“相当不错”的策略；而如果自车选择让行无避让，则可以保证驾驶安全，另外自车也不需要进行横向避让，该策略确保安全性，总体来说也是一个“相当不错”的策略。而对于抢行无避让的策略而言，这样虽然可以保证自车的通行的顺畅性，但是由于在横向没有避让动作，因此直行的自车和汇入的他车之间发生碰撞的几率较大，因此该策略的安全性较低，而安全性对于车辆驾驶而言是第一性的要求，因此该策略不是一个优选的方案；而对于让行避让策略，既然自车选择了让行，那么避让就是一个冗余的动作，因此该策略也不是一个优选的方案。所以综合以上四种决策方案子集的目标决策方案的考虑；抢行避让和让行无避让是较优选的方案；因此将它们作为第一梯队的策略方案集合；而抢行无避让和让行避让不是优选方案，因此将它们作为第二梯队的策略方案集合。上述的“优选方案”和“非优选方案”的判断基准是以决策收益和设定阈值进行比较而确定的。
应当理解的是：上述的第一梯队和第二梯队的策略集合是对于自车当前所处的状态而言的，随着自车和博弈目标的状态变换，各种策略之间的收益得分和相应的优先级随时可能发生变化。
继续上述图14所示场景及相关策略，当区分了第一梯队的目标决策方案集合和第二梯队的目标决策方案集合之后，规划模块进一步确认第一梯队的目标决策方案之间的决策收益差，当决策收益之间的差小于设定阈值时，认为当前第一梯队的目标决策方案集合之间暂无最优方案。例如，上述的第一梯队中包括抢行让行和让行无避让这两种目标决策方案，它们的决策收益分别为0.8和0.77，如果设定阈值为0.1，则这两种目标决策方案(抢行让行，让行无避让)之间的决策收益差(0.03)小于设定阈值，则规划模块认为在当前时刻，上述两个目标决策方案之间没有明显的优劣之分，自车采用哪一个方案都是可以的，则此时可以理解为自车与博弈目标的交互关系不明确，需要进入第一交互阶段，在该第一交互阶段中，保留自车与博弈目标在相对应的多个不同的决策方案之间跳转的可能性。
进一步的，规划模块在确定自车与博弈目标的交互满足进入第二交互阶段的条件时，由第一交互阶段进入第二交互阶段，此时自车与博弈目标的交互关系明确。
其中，规划模块可以基于下述条件来判断自车是否由第一交互阶段进入第二交互阶段。
条件一：规划模块可以基于目标决策方案集中决策收益第一高的目标决策方案与决策收益第二高的目标决策方案之间的决策收益差与阈值收益差之间的大小关系，确定是否满足进入第二交互阶段的条件。当决策方案集中决策收益第一高的决策方案与决策收益第二高的决策方案之间的决策收益差大于于阈值收益差时，规划模块确定满足进入第二交互阶段的条件。
条件二：规划模块可以基于博弈目标在预定时刻的实际运动轨迹与第一决策方案所预判的博弈目标的轨迹的比较结果，确定是否满足进入第二交互阶段的条件。
在条件二的一种情况下，当规划模块确定博弈目标在预定时刻的实际运动轨迹与第一决策方案预测的博弈目标在预定时刻的运动轨迹较为高度地一致时，规划模块确定满足进入第二交互阶段的条件。此时，第一交互阶段采用的第一决策方案与第二交互阶段采用的第二决策方案相同。
在条件二的另一种情况下，当规划模块确定博弈目标在预定时刻的实际运动轨迹与第一决策方案预测的博弈目标在预定时刻的运动轨迹完全不一致时，规划模块确定满足进入第二交互阶段的条件。可以理解的，当规划模块确定博弈目标在预定时刻的实际运动轨迹与目标决策方案(例如：上述的“抢行避让”)预测的博弈目标在预定时刻的运动轨迹完全不一致时，说明博弈目标的运动轨迹完全出乎预料，意味着当前用于博弈交互的决策方案必定不适用，此时，规划模块需要将当前用于博弈交互的目标决策方案A调整为更适合本次博弈的目标决策方案(例如，目标决策方案B)，决策方案B可以从目标决策方案集中获取，例如，如果当前自车发现抢行避让策略不再适用(博弈目标可能并未按照预想的让行而是无减速抢行)，那么可以在决策方案集中获取更适合的目标决策方案，例如让行避让，当确定了决策方案B后，规划模块可以确认当前自车与博弈目标的交互关系明确，进入第二交互阶段。
进一步的，规划模块还可以将条件一与条件二进行结合，共同用于确定是否满足进入第二交互阶段的条件。
当通过条件一与条件二进行结合，确定是否满足进入第二交互阶段的条件时，若基于条件一判断是否满足进入第二交互阶段的结果，与基于条件二判断是否满足进入第二交互阶段的结果不一致时，以基于条件而判断是否满足进入第二交互阶段的结果为主，即条件二的优先级高于条件一的优先级，这样设定的原因在于：需要注意，一致/不一致性的跳转通常是由于博弈目标没有按照交规或一般驾驶习惯决策，可能与代价的评估相违背，其优先级是高于基于收益跳转的。
可选的，该规划模块可以根据该博弈目标的横、纵向动作序列进行一致性或不一致性的判断。
其中，基于上述第二方面，规划模块确定博弈目标的预定时刻的预测轨迹是否与博弈目标预定时刻的真实轨迹相符时，规划模块对于博弈目标预定时刻的未来状态(即预测轨迹)可以基于持续的历史状态改变来判断，基于每一时选取的进行交互的决策方案中博弈目标预测的动作序列和该博弈目标实际的动作序列进行判断，其中，在进行判断过程中， 可以区分横向纵向。
示例性的，以博弈目标为机动车辆为例，利用统计数据处理方法，对期望加速度与实际加速度序列进行处理，得到如图17中(a)所示的纵向动作序列。
其中，根据上述图17中(a)所示的内容，可知a1～a3为规划模块确定的该博弈目标在预期时长内的期望加速度变化情况，b1～b3为该博弈目标当前时长内的实际加速度变化情况，b4～b6为该博弈目标在预期时长内的实际加速度变化情况，若a1～a3与b4～b6之间相差波动超过阈值范围内，则规划模块确定博弈目标在预定时刻的跟踪结果与第一决策方案预测的博弈目标在预定时刻的运动轨迹不一致；若a1～a3与b4～b6之间相差波动在阈值范围内，则规划模块确定博弈目标在预定时刻的跟踪结果与第一决策方案预测的博弈目标在预定时刻的运动轨迹一致。
示例性的，以博弈目标为单车为例，利用统计数据处理方法，对期望距离与实际距离序列进行处理，得到如图17中(b)所示的横向动作序列。其中，由于横向的动作的改变往往是不可逆的，对于横向一致/不一致的判断会受历史判断结果的影响。
其中，根据上述图17中(b)所示的内容，可知c1～c3为规划模块确定的该博弈目标在预期时长内的期望距离变化情况，d1～d3为博弈目标当前时长内的实际距离变化情况，d4～d6为博弈目标在该预期时长内的实际距离变化情况，若c1～c3与d4～d6之间相差波动超过阈值范围，则规划模块确定该博弈目标在预定时刻的跟踪结果与该第一决策方案预测的博弈目标在预定时刻的运动轨迹不一致；若c1～c3与d4～d6之间相差波动在阈值范围内，则规划模块确定该博弈目标在预定时刻的跟踪结果与该第一决策方案预测的博弈目标在预定时刻的运动轨迹一致。
综上，本申请可以根据实际情况将交互博弈的过程划分为多个交互阶段。
其中，第一交互阶段可以称为临界状态。临界状态可以理解为自车与博弈目标的交互关系不明确，即自车与博弈目标进行交互的决策方案可能随时发生改变。
其中，第二交互阶段可以称为置信状态。置信状态可以理解为自车与博弈目标的交互关系明确，即自车与博弈目标进行交互的决策方案确定。
例如，自车与博弈目标进行博弈的过程中，随着博弈过程的继续，自车会趋近于一个决策方案，当趋近的决策方案的决策收益与其他决策方案的决策收益的差值大于阈值收益差时，认为博弈关系明确，完成从临界态到置信态的跳转。
图18示出了上述的交互博弈的整体流程逻辑。
以上所描述的装置实施例仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。另外，本申请提供的装置实施例附图中，模块之间的连接关系表示它们之间具有通信连接，具体可以实现为一条或多条通信总线或信号线。
通过以上的实施方式的描述，所属领域的技术人员可以清楚地了解到本申请可借助软件加必需的通用硬件的方式来实现，当然也可以通过专用硬件包括专用集成电路、专用CPU、专用存储器、专用元器件等来实现。一般情况下，凡由计算机程序完成的功能都可以很容易地用相应的硬件来实现，而且，用来实现同一功能的具体硬件结构 也可以是多种多样的，例如模拟电路、数字电路或专用电路等。但是，对本申请而言更多情况下软件程序实现是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在可读取的存储介质中，如计算机的软盘、U盘、移动硬盘、ROM、RAM、磁碟或者光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。
所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本申请实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存储的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质(例如固态硬盘(Solid State Disk，SSD))等。
尽管在此结合各实施例对本发明进行了描述，然而，在实施所要求保护的本发明过程中，本领域技术人员通过查看所述附图、公开内容、以及所附权利要求书，可理解并实现所述公开实施例的其它变化。在权利要求中，“包括”(comprising)一词不排除其他组成部分或步骤，“一”或“一个”不排除多个的情况。单个处理器或其它单元可以实现权利要求中列举的若干项功能。相互不同的从属权利要求中记载了某些措施，但这并不表示这些措施不能组合起来产生良好的效果。
以上所述，仅为本申请的具体实施方式，但本申请的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请的保护范围之内。因此，本申请的保护范围应以所述权利要求的保护范围为准。

Claims (13)

  1. 一种自动驾驶装置，包括：
    处理装置，所述处理装置包括多个处理器核；
    所述多个处理器核中的至少一部分处理器核被固定分配给设定的任务集合，所述固定分配指：所述多个处理器核中的至少一部分处理器核被配置为仅处理设定的任务集合，所述设定的任务集合为非空集合。
  2. 根据权利要求1所述的装置，还包括：
    用于获取自车感知信息的感知系统，所述感知系统包括以下至少一种：激光雷达，毫米波雷达，摄像头，超声波雷达；
    所述设定的任务集合包括感知任务，所述感知任务包括由所述至少一部分处理器核处理由所述感知系统所获取的信息并获取车辆周围的环境信息和障碍物信息。
  3. 根据权利要求2所述的装置，其特征在于：
    所述障碍物信息包括以下至少一种：静态障碍物信息，动态障碍物信息。
  4. 根据权利要求1-3任一所述的装置，还包括：
    规划系统；
    所述设定的任务集合包括规划任务，所述规划任务包括由所述的至少一部分处理器核基于所述车辆的运动信息核所述障碍物信息确定所述车辆的规划行驶策略。
  5. 根据权利要求4所述的装置，其特征在于：
    所述感知任务包括以下至少一种：用于识别车辆周围车道线的车道线识别(LD)任务，用于识别车辆周围其它车辆的车辆检测(VD)，用于将不同传感器信息进行融合的感知融合(RL)，对车辆周围的关键目标进行感知和跟踪，所述关键目标包括弱势道路使用者(VRU)；
    所述规划任务包括以下至少一种：车辆的全局路径规划(Router)，有限状态机(FSM)，运动控制(MOP)。
  6. 根据权利要求5所述的装置，其特征在于：
    将所述至少一部分处理器核中的至少一个处理器核固定分配给所述感知任务中的一种任务。
  7. 根据权利要求5所述的装置，其特征在于：
    将所述至少一部分处理器核中的至少一个处理器核固定分配给所述规划任务中的一种任务。
  8. 根据权利要求6所述的装置，其特征在于：
    将所述至少一部分处理器核中的至少一个处理器核固定分配给对车辆周围的关键 目标进行感知和跟踪，所述关键目标包括弱势道路使用者(VRU)。
  9. 根据权利要求1-8任意一项所述的装置，还包括
    根据所述感知信息确定至少一个目标；
    获取所述至少一个目标的未来运动轨迹；
    获取所述自车的未来运动轨迹；
    将所述至少一个目标中未来运动轨迹与所述自车的未来运动轨迹有交叉的目标确定为所述博弈目标。
  10. 根据权利要求9所述的装置，还包括：
    根据感知信息获取自车的决策方案集；
    从所述决策方案集中获取用于所述自车与博弈目标进行第一交互阶段的第一决策方案；
    在基于所述第一决策方案控制所述自车进行行驶的过程中，确定所述自车与所述博弈目标满足进入第二交互阶段的条件；
    从所述决策方案集中获取用于所述自车与所述博弈目标进行第二交互阶段的第二决策方案；
    基于所述第二决策方案控制所述自车进行行驶。
  11. 根据权利要求10所述的装置，其中：
    所述第一交互阶段为自车与所述博弈目标的交互关系不明确的阶段；所述第二交互阶段为自车于所述博弈目标的交互关系明确的阶段。
  12. 根据权利要求9-11任一所述的装置，还包括：
    在所述自车与所述博弈目标进行所述第一交互阶段的过程中，对所述博弈目标的状态进行持续跟踪。
  13. 一种车辆，包括：
    如权利要求1-12任一所述的装置。
PCT/CN2022/076331 2022-02-15 2022-02-15 一种智能驾驶方法、装置及包括该装置的车辆 WO2023155041A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/076331 WO2023155041A1 (zh) 2022-02-15 2022-02-15 一种智能驾驶方法、装置及包括该装置的车辆

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/076331 WO2023155041A1 (zh) 2022-02-15 2022-02-15 一种智能驾驶方法、装置及包括该装置的车辆

Publications (1)

Publication Number Publication Date
WO2023155041A1 true WO2023155041A1 (zh) 2023-08-24

Family

ID=87577234

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/076331 WO2023155041A1 (zh) 2022-02-15 2022-02-15 一种智能驾驶方法、装置及包括该装置的车辆

Country Status (1)

Country Link
WO (1) WO2023155041A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117246320A (zh) * 2023-11-10 2023-12-19 新石器慧通（北京）科技有限公司 车辆的控制方法、装置、设备及存储介质
CN118424321A (zh) * 2024-07-05 2024-08-02 新石器慧通（北京）科技有限公司 自动驾驶车辆的轨迹规划方法、装置、设备及车辆

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105955806A (zh) * 2015-03-09 2016-09-21 佳能株式会社 程序处理设备和方法以及图像形成设备
CN108225364A (zh) * 2018-01-04 2018-06-29 吉林大学 一种无人驾驶汽车驾驶任务决策系统及方法
CN108734300A (zh) * 2017-04-24 2018-11-02 英特尔公司 使用自主机器的识别、再标识和安全性增强
CN110689115A (zh) * 2019-09-24 2020-01-14 上海寒武纪信息科技有限公司 神经网络模型处理方法、装置、计算机设备及存储介质
CN111831434A (zh) * 2020-07-01 2020-10-27 Oppo广东移动通信有限公司 资源分配方法、装置、存储介质及电子设备
CN112373485A (zh) * 2020-11-03 2021-02-19 南京航空航天大学 一种考虑交互博弈的自动驾驶车辆决策规划方法
CN113276884A (zh) * 2021-04-28 2021-08-20 吉林大学 一种可变博弈模式的智能车交互决策通行方法及系统
US20220032960A1 (en) * 2020-07-29 2022-02-03 Toyota Research Institute, Inc. Game-theoretic planning for risk-aware interactive agents

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105955806A (zh) * 2015-03-09 2016-09-21 佳能株式会社 程序处理设备和方法以及图像形成设备
CN108734300A (zh) * 2017-04-24 2018-11-02 英特尔公司 使用自主机器的识别、再标识和安全性增强
CN108225364A (zh) * 2018-01-04 2018-06-29 吉林大学 一种无人驾驶汽车驾驶任务决策系统及方法
CN110689115A (zh) * 2019-09-24 2020-01-14 上海寒武纪信息科技有限公司 神经网络模型处理方法、装置、计算机设备及存储介质
CN111831434A (zh) * 2020-07-01 2020-10-27 Oppo广东移动通信有限公司 资源分配方法、装置、存储介质及电子设备
US20220032960A1 (en) * 2020-07-29 2022-02-03 Toyota Research Institute, Inc. Game-theoretic planning for risk-aware interactive agents
CN112373485A (zh) * 2020-11-03 2021-02-19 南京航空航天大学 一种考虑交互博弈的自动驾驶车辆决策规划方法
CN113276884A (zh) * 2021-04-28 2021-08-20 吉林大学 一种可变博弈模式的智能车交互决策通行方法及系统

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117246320A (zh) * 2023-11-10 2023-12-19 新石器慧通（北京）科技有限公司 车辆的控制方法、装置、设备及存储介质
CN117246320B (zh) * 2023-11-10 2024-02-09 新石器慧通（北京）科技有限公司 车辆的控制方法、装置、设备及存储介质
CN118424321A (zh) * 2024-07-05 2024-08-02 新石器慧通（北京）科技有限公司 自动驾驶车辆的轨迹规划方法、装置、设备及车辆

Similar Documents

Publication Publication Date Title
US11860639B2 (en) Vehicle-based road obstacle identification system
WO2021136130A1 (zh) 一种轨迹规划方法及装置
CN110775063B (zh) 一种车载设备的信息显示方法、装置及车辆
WO2021135371A1 (zh) 一种自动驾驶方法、相关设备及计算机可读存储介质
US11230285B2 (en) System and method for determining a vehicle action
US11945440B2 (en) Data driven rule books
WO2021147748A1 (zh) 一种自动驾驶方法及相关设备
CN112859830B (zh) 一种设计运行区域odd判断方法、装置及相关设备
WO2021098222A1 (zh) 一种车辆行驶控制方法及相关设备
WO2019080874A1 (zh) 一种车辆借道通行的方法和控制中心
US11520342B2 (en) System and method for determining realistic trajectories
US20210191394A1 (en) Systems and methods for presenting curated autonomy-system information of a vehicle
WO2023155041A1 (zh) 一种智能驾驶方法、装置及包括该装置的车辆
CN113968216A (zh) 一种车辆碰撞检测方法、装置及计算机可读存储介质
WO2022205243A1 (zh) 一种变道区域获取方法以及装置
CN112672942A (zh) 一种车辆换道方法及相关设备
CN114964286A (zh) 轨迹规划信息生成方法、装置、电子设备以及存储介质
Choi et al. Framework for connected and automated bus rapid transit with sectionalized speed guidance based on deep reinforcement learning: Field test in Sejong city
JP2023156266A (ja) シミュレーション方法、コンピューティングデバイス及び記憶媒体
US20230256999A1 (en) Simulation of imminent crash to minimize damage involving an autonomous vehicle
CN112829762A (zh) 一种车辆行驶速度生成方法以及相关设备
WO2021254000A1 (zh) 车辆纵向运动参数的规划方法和装置
WO2020139392A1 (en) Vehicle-based road obstacle identification system
WO2023102827A1 (zh) 一种路径约束方法及装置
CN113963535B (zh) 行驶决策确定方法、装置、电子设备存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22926379

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 2022926379

Country of ref document: EP

ENP Entry into the national phase

Ref document number: 2022926379

Country of ref document: EP

Effective date: 20240808

NENP Non-entry into the national phase

Ref country code: DE