RU2805539C2 - Method and device for controlling an unmanned vehicle - Google Patents

Abstract

FIELD: unmanned vehicles.

SUBSTANCE: sensor system is installed on an unmanned vehicle and collects information about the environment. The unmanned vehicle is connected to an electronic device. The method includes the steps at which at a given time during operation of an unmanned vehicle on a road section: generating, ranking and classifying the first and second potential trajectories; identifying the best and safest path among the first and second potential paths; the first or second action is ordered to perform. The ranking algorithm indicates the similarity between the corresponding potential and reference trajectories. The classifying algorithm generates a security score. The best and safest potential trajectories are identified based on a combination of appropriate ranking scores and safety scores. The first action is taken if the best trajectory and the safest trajectory are the same potential trajectory. The second action is taken if the best trajectory and the safest trajectory are not the same potential trajectory.

EFFECT: increased safety of manoeuvre while preventing collision of an unmanned vehicle with other objects.

22 cl, 17 dwg

Description

Technical field

[0001] The present technology generally relates to a self-driving car; and, in particular, to a method and device for controlling an unmanned vehicle.

State of the art

[0001] Several computer navigation systems that are configured to assist in navigation and/or control of a vehicle have been proposed and implemented in the prior art. These systems range from simpler solutions based on map localization, e.g. Using a computer system to assist the driver in navigating a route from a starting point to an ending point; to more complex ones - computer and/or driver-autonomous driving systems.

[0002] Some of these systems are implemented as so-called "cruise control" systems. In these systems, a computer system installed on the vehicles maintains the vehicle speed set by the user. Some cruise control systems implement an "intelligent distance control" system whereby the user can set the distance to a potential vehicle ahead (for example, select a value expressed in the number of vehicles) and the computer system adjusts the vehicle's speed, at least in part, by based on a vehicle approaching a potential vehicle ahead within a predetermined distance. Some cruise control systems are additionally equipped with a collision monitoring system that, when it detects a vehicle (or other obstacle) in front of a moving vehicle, slows or stops the vehicle.

[0003] Some of the more advanced systems provide fully autonomous driving of a vehicle without direct control by an operator (i.e., driver), so-called self-driving vehicles (SDC). This SDC includes computer systems that can instruct the SDC to accelerate, brake, stop, change lanes, and park itself.

[0004] A secure electronic system may require planning a smooth as well as a safe trajectory for the SDC, taking into account the surroundings. The SDC environment can be very dynamic because it may include objects (eg, other vehicles, including other two-wheelers, four-wheelers or multi-wheelers, a person walking on the road, etc.) that may already be in motion or start moving when they are near the SDC.

[0005] For example, CN 111332298 A discloses a method, apparatus and equipment for determining a driving accessible area and an information carrier, and relates to the field of automatic driving. According to a specific implementation scheme, the method is applied to electronic equipment, the sensor system is carried on the target vehicle, and the electronic equipment interacts with the sensor system. The method comprises the steps of collecting environmental data of the current frame collected by the sensor system of the target vehicle; if it is determined that an adjacent lane exists for the current lane of the target vehicle, according to the environment data of the current frame, extracting environmental feature data of the current frame in the adjacent lane; predicting the hazard of the current frame of the adjacent lane according to the environmental characteristic data of the current frame to obtain an initial hazard value of the current frame; calculating the final hazard value of the current frame according to the initial hazard value of the current frame and the corresponding reference hazard value of the previous frame; and calculating the drivable area of the current frame of the adjacent lane according to the final hazard value of the current frame. Therefore, the target vehicle cannot misdirect too much, and the stability of the target vehicle and the user's driving experience are improved.

[0006] "US 2020/0310428 A1" discloses a lane changing method, an apparatus for an autonomous vehicle, and a computer-readable storage medium. The lane changing method includes: determining possible lanes to which a lane change should be performed based on the driving intention of a host vehicle; checking possible lanes based on lane change conditions for possible lanes; selecting from the verified possible lanes the target lane to which the main vehicle's lane change should be made; determining whether the host vehicle's lane change from the current lane to the target lane is safe; and performing a lane change if the lane change is determined to be safe.

[0007] US 10,163,353 B2 discloses a control system that is adapted for use in a vehicle and is designed to detect following vehicles based on environmental data that is obtained from one or more environmental sensors located on the vehicle. The environmental sensors are adapted to provide the electronic control system with environmental data that reflects the area in front of, next to and/or behind the vehicle. The control system is at least adapted and configured to detect one or more other vehicles involved in traffic behind its own vehicle using environmental sensors. The lane associated with every other vehicle in which the other vehicle(s) is traveling is detected. The traffic density of one's own lane and/or at least one adjacent lane of other vehicle(s) is determined. Based on the determined traffic density, (i) an indication of a safe or unsafe lane change to a specific one of at least one adjacent lane is provided and/or (ii) an autonomous lane change to a specific one of at least one adjacent lane is performed by the vehicle's own means if said specified traffic density allows for a safe lane change.

[0008] US Pat. No. 10,513,261 B2 discloses a method for determining the potential hazard of an own vehicle changing lanes from the currently used lane to an adjacent lane, estimating the detection range in the external space of the monitored own vehicle, and the impact of objects identified in the detection range on the potential hazard, and based on the positions and speeds of internal other vehicles identified within the detection range, it is determined whether external other vehicles located outside the detection range can reach the target area in which the own vehicle is located after the intended lane change. A method for at least partially automated control of one's own vehicle, wherein, in the event of an intended lane change, the potential danger of that lane change is assessed and the lane change is prevented if other external vehicles can reach the target area.

[0009] One of the technical challenges in implementing the above-mentioned electronic systems is to perform a smooth as well as safe maneuver while preventing any collision of the SDC with other objects.

The essence of the invention

[00010] The objective of the present technology is to eliminate at least some of the inconveniences present in the prior art. Embodiments of the present technology may provide and/or expand the scope of approaches to achieving the goals and objectives of the present technology.

[00011] A self-driving vehicle (SDC) is typically equipped with an electronic device and various sensors. For example, the electronic device may be configured to receive and process data about the lanes of a road segment along which the SDC is traveling, such as data indicating the geometry, shape, and boundaries of the lanes. In addition, the electronic device may be configured to receive and process data about objects in the SDC environment, such as, for example, data indicative of object types, object movement, and object geometry.

[00012] The developers of the present technology have realized that when SDC is operating on a stretch of road, the SDC may require various lane change maneuvers to be performed. The path used by the SDC when performing a lane change maneuver must be safe to avoid any collision between the SDC and other objects. In addition, the trajectory should provide a smooth and comfortable maneuver.

[00013] Developers of the present technology have an electronic device and a method performed on the electronic device to determine a path that is safe and also allows for smooth and comfortable maneuvering. To achieve this, the electronic device can generate potential trajectories based on processed data about traffic lanes and objects in the SDC environment. The electronic device can independently rank and classify potential trajectories based on ranking scores and safety scores. The ranking scores may be related to the similarity of the predicted trajectory to a reference trajectory, and the reference trajectory may be manually performed by a human expert. In some examples, the reference trajectory can provide a smooth and comfortable maneuver. Safety assessments can indicate whether a corresponding trajectory is safe.

[00014] The electronic device may identify the best trajectory based on the ranking scores and the safest trajectory based on the safety assessment. In the event that the best trajectory and the safest trajectory coincide with the potential trajectory, the electronic device may direct the SDC to perform a maneuver in accordance with the same potential trajectory. If the best trajectory and the safest trajectory do not coincide with the potential trajectory, the electronic device may direct the SDC to perform the maneuver in accordance with the safest trajectory. In addition, if the best trajectory and safest trajectory do not coincide with the potential trajectory, the electronic device may perform other actions, such as turning on the visual SDC indicator to indicate the intention to perform a lane change maneuver.

[00015] In a first broad aspect of the present technology, a method is provided for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's environment, wherein the SDC is coupled to an electronic device, the method including the steps of: of which: at a given point in time while the SDC is operating on the road section: generating, using an electronic device, a first potential trajectory and a second potential trajectory for the SDC on the road section, respectively; ranking, by an electronic device using a machine learning ranking algorithm (MLA), the first candidate trajectory and the second candidate trajectory, wherein the ranking MLA is trained to predict a ranking score indicating the similarity between the corresponding candidate trajectory and the reference trajectory; classifying, by an electronic device using a classifying MLA, a first candidate trajectory and a second potential trajectory, the classifying MLA being trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; electronically identifying the best trajectory among the first potential trajectory and the second potential trajectory and the safest trajectory among the first potential trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; direct the electronic SDC device to perform the first action if the best trajectory and the safest trajectory are the same potential trajectory; and direct the electronic SDC device to perform a second action if the best trajectory and the safest trajectory are not the same potential trajectory.

[00016] In some embodiments of the method, the ranking MLA is a pairwise ranking algorithm and is trained on a first set of training data, the first set of training data includes: a reference trajectory, which is a training trajectory manually executed by a human expert moving in the SDC on section of the road for training at the time of training; a first predicted training trajectory and a second predicted training trajectory generated by the electronic device for the SDC for the training road section at the training time; and the ranking MLA is trained on the first training data set to rank the first predicted training trajectory and the second predicted training trajectory for the training road segment based on corresponding similarity to the reference trajectory.

[00017] In some embodiments of the method, the classifying MLA is trained based on a second set of training data, the second set of training data comprising: a training trajectory generated by an electronic SDC device at a training road segment during a training time point; a label indicating at least one of the implicit and explicit inputs of a human expert moving in the SDC on the training road segment during the training time point.

[00018] In some embodiments of the method, the classifying MLA is a decision tree-based algorithm.

[00019] In some embodiments of the method, a section of road has two lanes, and the first potential trajectory includes a potential lane change maneuver, and the second potential trajectory includes a potential lane keeping maneuver, and wherein: if both the best trajectory and the safest the trajectory is a first potential trajectory, the first action is an attempt to perform a potential lane change maneuver based on the first potential trajectory; if both the best trajectory and the safest trajectory are the second potential trajectory, the first action is to attempt a potential lane keeping maneuver based on the second potential trajectory.

[00020] In some embodiments of the method, a section of road has two lanes, and the first potential trajectory includes a potential lane change maneuver, and the second potential trajectory includes a potential lane keeping maneuver, and wherein: if the best trajectory is the first potential trajectory, and the safest trajectory is the second potential trajectory, the second action is to turn on the SDC visual indicator to indicate the intention to perform a potential lane change maneuver when performing a potential lane keeping maneuver until another predetermined time point after a given time point.

[00021] In some embodiments of the method, a section of road has two lanes, and the first potential trajectory includes a potential lane change maneuver, and the second potential trajectory includes a potential lane keeping maneuver, and wherein: if the best trajectory is the second potential trajectory, and the safest trajectory is the first potential trajectory, the second action is to attempt a potential lane change maneuver based on the first potential trajectory.

[00022] In some embodiments of the method, the method further includes the steps of: at another predetermined time after the predetermined time: electronically identifying a new best path among the first potential path and the second potential path, as well as the new safest a trajectory among the first and second potential trajectories based on a combination of the respective new ranking scores and the new safety scores; in response to the new best trajectory and the new safest trajectory, which is the first potential trajectory, causing the SDC electronic device to perform a potential lane change maneuver based on the first potential trajectory.

[00023] In a second broad aspect of the present technology, a method is provided for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's environment, wherein the SDC is coupled to an electronic device, the method including the steps of: of which: at a given point in time while the SDC is operating on a road section: ranking by means of an electronic device using a ranking MLA a first potential trajectory and a second potential trajectory, the ranking MLA being trained to predict a ranking score indicating the similarity between the corresponding potential trajectory and the reference trajectory, without taking into account the safety of the corresponding potential trajectory; classifying, by an electronic device using a classifying MLA, a first candidate trajectory and a second potential trajectory, the classifying MLA being trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; causing the electronic SDC device to perform an action based on a combination of the ranking scores and the safety scores for the first candidate trajectory and the second candidate trajectory.

[00024] In some embodiments of the method, the classification of the first and second potential trajectories is performed in parallel with the ranking of the first potential trajectory and the second potential trajectory.

[00025] In some embodiments of the method, the classification of the first and second candidate trajectories is performed independently of the ranking of the first candidate trajectory and the second potential trajectory.

[00026] In some embodiments of the method, a section of road has two lanes of traffic, and the first potential trajectory includes a potential lane change maneuver in response to a determination by an electronic device that the first potential trajectory ranks higher than the second potential trajectory and that the first trajectory is safe, wherein the injunction includes the step of: instructing the electronic SDC device to attempt to perform a potential lane change maneuver based on the first potential trajectory.

[00027] In a third broad aspect of the present technology, an electronic device is provided for controlling a driverless vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's surroundings, wherein the SDC is coupled to the electronic device, wherein the electronic device is configured to : at a given time during the operation of the SDC on the road section: generate the first potential trajectory and the second potential trajectory for the SDC on the road section, respectively; rank, using a ranking machine learning algorithm (MLA), the first candidate trajectory and the second candidate trajectory, the ranking MLA being trained to predict a ranking score indicating similarity between the corresponding candidate trajectory and the reference trajectory; classify, using the classifying MLA, the first candidate trajectory and the second potential trajectory, the classifying MLA being trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; identify the best trajectory among the first candidate trajectory and the second candidate trajectory and the safest trajectory among the first candidate trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; instruct the SDC to take the first action if the best trajectory and the safest trajectory are the same potential trajectory; and direct the SDC to take a second action if the best trajectory and the safest trajectory are not the same potential trajectory.

[00028] In some embodiments of an electronic device, the ranking MLA is a pairwise ranking algorithm and is trained on a first set of training data, wherein the first set of training data includes: a reference trajectory, which is a training trajectory manually executed by a human expert moving in SDC on the road section for training at the training time; and a first predicted training trajectory and a second predicted training trajectory generated by the electronic device for the SDC for the training road section at the training time; and the ranking MLA is trained on the first training data set to rank the first predicted training trajectory and the second predicted training trajectory based on the similarity between (i) the first predicted training trajectory and the second predicted training trajectory and (ii) the reference trajectory.

[00029] In some embodiments of the electronic device, the classifying MLA is trained based on a second set of training data, the second set of training data comprising: a training trajectory generated by the SDC electronic device at a training road segment during a training time point; a label indicating at least one of the implicit and explicit inputs of a human expert driving in the SDC on the training road section during the training time point.

[00030] In some embodiments of an electronic device, the classifying MLA is a decision tree-based algorithm.

[00031] In some embodiments of an electronic device, a stretch of road has two lanes of traffic, and the first potential trajectory includes a potential lane change maneuver, and the second potential trajectory includes a potential lane keeping maneuver, wherein: if both the best trajectory and the safest trajectory is the first potential trajectory, the first action is to attempt a potential lane change maneuver based on the first potential trajectory; if both the best trajectory and the safest trajectory are the second potential trajectory, the first action is to attempt a potential lane keeping maneuver based on the second potential trajectory.

[00032] In some embodiments of an electronic device, a stretch of road has two lanes of traffic, and the first potential trajectory comprises a potential lane change maneuver, and the second potential trajectory comprises a potential lane keeping maneuver, and wherein: if the best trajectory is the first potential trajectory and the safest trajectory is the second potential trajectory, the second action is to turn on the SDC visual indicator to indicate the intention to perform the potential lane change maneuver when performing the potential lane keeping maneuver until another predetermined time after the predetermined time.

[00033] In some embodiments of an electronic device, a stretch of road has two lanes of traffic, and the first potential trajectory comprises a potential lane change maneuver, and the second potential trajectory comprises a potential lane keeping maneuver, and wherein: if the best trajectory is the second potential trajectory and the safest trajectory is the first potential trajectory, the second action is to attempt a potential lane change maneuver based on the first potential trajectory.

[00034] In some embodiments of the electronic device, the electronic device is further configured to: at another given point in time after a given point in time: identify a new best trajectory among the first potential trajectory and the second potential trajectory and a new safest trajectory among the first potential trajectory and the second potential trajectories based on a combination of the corresponding new ranking scores and new safety scores; in response to both the new best trajectory and the new safest trajectory being the first potential trajectory, causing the SDC to perform a potential lane change maneuver based on the first potential trajectory; in response to the fact that both the new best trajectory and the new safest trajectory are the second potential trajectory, cause the SDC to perform a potential lane keeping maneuver based on the second potential trajectory.

[00035] In a fourth broad aspect of the present technology, a method is provided for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's surroundings, wherein the SDC is coupled to an electronic device, the method including the steps of: at which: at a given time during the operation of the SDC on the road section: a first potential trajectory and a second potential trajectory for the SDC on the road section are generated, respectively, using an electronic device; electronically ranking the first candidate trajectory and the second candidate trajectory using a ranking machine learning algorithm (MLA), wherein the ranking MLA is trained to predict a ranking score indicating similarity between the corresponding candidate trajectory and the reference trajectory; classifying, by an electronic device using a classifying MLA, a first candidate trajectory and a second potential trajectory, the classifying MLA being trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; electronically identifying the best trajectory among the first potential trajectory and the second potential trajectory and the safest trajectory among the first potential trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; order, using the electronic SDC device, to work in accordance with the safest trajectory; and if the safest path and the best path are not the same potential path, cause the SDC electronic device to perform a corrective action by operating in accordance with the safest path, wherein the corrective action causes the best path to become the safest path in the future.

[00036] In some embodiments of the method, the corrective action is to turn on the visual SDC indicator to indicate the intent to perform a potential lane change maneuver.

[00037] As used herein, the term "environment" of a given vehicle refers to the area or volume around a given vehicle, including the portion of its current environment available for scanning using one or more sensors mounted on a given vehicle, e.g. a three-dimensional map of such an environment or detection of objects in it.

[00038] As used herein, a "server" is a computer program that runs on suitable hardware and is capable of receiving requests (eg, from electronic devices) over a network and executing or causing those requests to be executed. completed. The hardware may be implemented as one physical computer or one physical computer system, but neither is required for the present technology. In this context, the use of the expression "server" is not intended to indicate that every task (such as received instructions or requests) or any particular task was accepted, executed, or called to execution by the same server (that is, the same software). software and/or hardware); this means that any number of software elements or hardware devices may be involved in receiving/sending, executing or providing the performance of any task or request or the consequences of any task or request; and all of this software and hardware may be a single server or multiple servers, both of which are included in the expression “at least one server.”

[00039] As used herein, an "electronic device" is any computer hardware that is capable of running software suitable for the task at hand. As used herein, the term "electronic device" implies that the device can function as a server for other electronic devices, but this is not necessarily the case with respect to the present technology. Thus, some (non-limiting) examples of electronic devices include drones, personal computers (desktops, laptops, netbooks, etc.), smartphones and tablets, and network equipment such as routers, switches, and gateways. It should be understood that in this context, the fact that a device functions as an electronic device does not mean that it cannot function as a server for other electronic devices.

[00040] As used herein, the expression “information” includes information of any nature or kind that may be stored in a database. Thus, information includes, but is not limited to, visual works (for example, maps), audio-visual works (for example, images, films, sound recordings, presentations, etc.), data (for example, location data, weather data, traffic data, numerical data, etc.), text (eg opinions, comments, questions, messages, etc.), documents, spreadsheets, etc.

[00041] As used herein, the expression “software component” is intended to include software (suitable for a particular hardware context) that is both necessary and sufficient to perform the particular function(s) being referenced.

[00042] As used herein, the expression “computer storage medium” (also referred to as “storage medium”) is intended to include media of any nature or type, including, but not limited to, RAM, ROM, disks (CD-ROM, DVD, floppy disks, hard drives, etc.), USB keys, solid state drives, tape drives, etc. A plurality of components may be combined to form computer storage media, including two or more media components of the same type and/or two or more media components of different types.

[00043] As used herein, a “database” is any structured collection of data, regardless of its specific structure, database management software, or computer hardware on which the data is stored, implemented, or otherwise made available for use. The database may reside on the same hardware as the process that stores or uses the information stored in the database, or it may reside on separate hardware, such as a dedicated server or multiple servers.

[00044] As used herein, the words "first", "second", "third", etc. were used as adjectives only for the purpose of providing a distinction between the nouns they modify relative to each other, and not for the purpose of describing any specific relationship between those nouns. Moreover, as discussed here in other contexts, reference to a "first" element and a "second" element does not prevent the two elements from being the same actual real-world element.

[00045] Each of the implementations of the present technology has at least one of the above objects and/or aspects, but does not necessarily have all of them. It should be understood that some aspects of the present technology that have arisen in an attempt to achieve the above purpose may not satisfy this purpose and/or satisfy other purposes that are not expressly described herein.

[00046] Additional and/or alternative features, aspects and advantages of implementations of the present technology will become apparent from the following description, the accompanying drawings and the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

[00047] These and other features, aspects and advantages of the present technology will become more apparent from the following description, the appended claims and the accompanying drawings, in which:

[00048] FIG. 1 depicts a schematic diagram of an exemplary computer system configured to implement certain non-limiting embodiments of the present technology;

[00049] FIG. 2 depicts a schematic diagram of a networked computing environment suitable for use with certain non-limiting embodiments of the present technology;

[00050] FIG. 3 is a schematic diagram of an electronic device for implementing various non-limiting embodiments of the present technology;

[00051] FIG. 4 illustrates a high-level functional block diagram of various modules included in a training module, in accordance with various non-limiting embodiments of the present technology;

[00052] FIG. 5 depicts a typical road section for training in accordance with various non-limiting embodiments of the present technology;

[00053] FIG. 6 depicts a typical timeline including training times in accordance with various non-limiting embodiments of the present technology;

[00054] FIG. 7 depicts another view of a road section for training in accordance with various non-limiting embodiments of the present technology;

[00055] FIG. 8 depicts another exemplary timeline including training times in accordance with various non-limiting embodiments of the present technology;

[00056] FIG. 9 depicts training a machine learning ranking algorithm (MLA) using a first set of training data as input in accordance with various non-limiting embodiments of the present technology;

[00057] FIG. 10 depicts training a classifying MLA using a second set of training data as input in accordance with various non-limiting embodiments of the present technology;

[00058] FIG. 11 illustrates a use case where SDC operates on a road section, in accordance with various non-limiting embodiments of the present technology;

[00059] FIG. 12 illustrates a trained ranking MLA trained to rank candidate trajectories in accordance with various non-limiting embodiments of the present technology;

[00060] FIG. 13 illustrates a trained classifier MLA trained to classify potential trajectories in accordance with various non-limiting embodiments of the present technology;

[00061] FIG. 14 illustrates a path identification module implemented as a submodule in a motion planning module, in accordance with various non-limiting embodiments of the present technology;

[00062] FIG. 15 illustrates a schematic diagram of a state transition mechanism implemented in a motion planning module, in accordance with various non-limiting embodiments of the present technology;

[00063] FIG. 16 illustrates a typical SDC operating scenario on a road segment where the state transition mechanism may be in a state of preparation for a lane change maneuver, in accordance with various non-limiting embodiments of the present technology; And

[00064] FIG. 17 illustrates a flow diagram of a method of operating an SDC in accordance with various non-limiting embodiments of the present technology.

Detailed description

[00065] The examples and conditional statements provided herein are intended primarily to assist the reader in understanding the principles of the present technology and not to limit its scope to such specific examples and conditions. It should be understood that those skilled in the art will be able to develop various mechanisms that, although not explicitly described herein, nevertheless embody the principles of the present technology and are included in the spirit and scope thereof.

[00066] In addition, the following description may describe implementations of the present technology in a relatively simplified form for purposes of ease of understanding. Those skilled in the art will appreciate that various implementations of the present technology can be highly complex.

[00067] In some cases, examples of modifications to the present technology that are considered useful may also be set forth. This is intended only to facilitate understanding and, again, not to strictly define the scope or delineate the boundaries of the present technology. These modifications are not an exhaustive list, and other modifications may be made by one skilled in the art while still remaining within the scope of the present technology. In addition, where examples of modifications are not provided, it should not be construed to mean that no modifications can be made and/or that what is described is the only way to implement such element of the present technology.

[00068] Moreover, all statements herein setting forth the principles, aspects and implementations of the technology, as well as specific examples thereof, are intended to cover both their structural and functional equivalents, whether currently known or under development. future. Thus, for example, those skilled in the art will recognize that any flowcharts herein represent conceptual views of an illustrative circuit embodying the principles of the present technology.

[00069] Likewise, it will be understood that any flowcharts, flowcharts, state transition diagrams, pseudo-codes, and the like represent various processes that may be substantially represented on computer-readable media and executed by a computer or processor regardless of whether such a computer or processor is explicitly shown or not.

[00070] The functions of the various elements shown in the figures, including any functional unit designated as a “processor,” may be provided through the use of dedicated hardware, as well as hardware capable of executing software in combination with associated software. When provided by a processor, the functions may be provided by one dedicated processor, one shared processor, or a plurality of separate processors, some of which may be shared.

[00071] Additionally, explicit use of the term "processor" or "controller" should not be construed as referring solely to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor (DSP) hardware ), a network processor, an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), read-only memory (ROM) for software storage, random access memory (RAM), and non-volatile storage. Other hardware, traditional and/or custom, may also be included.

[00072] Software modules, or simply modules, which may be understood as software, may be represented herein as any combination of flowchart elements or other elements indicating the execution of process steps and/or textual description. Such modules may be executed by hardware that is explicitly or implicitly shown.

[00073] With these fundamentals in mind, let us consider some non-limiting examples to illustrate various implementations of aspects of the present technology.

Computer system

[00074] Referring initially to FIG. 1, a computer system 100 is depicted suitable for use with certain implementations of the present technology, the computer system 100 includes various hardware components, including one or more single-core or multi-core processors collectively represented by a processor 110, a solid state drive 120, memory 130, which may be RAM or any other type of memory. Communication between the various components of the computer system 100 may be provided by one or more internal and/or external buses (not shown) (e.g., a PCI bus, a universal serial bus, an IEEE 1394 Firewire bus, a SCSI bus, a Serial-ATA bus, etc. ) to which the various components of the equipment are electronically coupled.

[00075] In at least some embodiments of the present technology, solid state drive 120 stores software instructions suitable for loading into memory 130 and execution by processor 110 to determine the presence of an object. object. For example, the software instructions may be part of a vehicle control application executed by processor 110.

[00076] In at least some embodiments of the present technology, it is contemplated that the computer system 100 may have additional and/or optional components, such as a network communications module 140 for communicating through a communications network (e.g., communications network 240 depicted in FIG. 2) with other electronic devices and/or servers, localization modules (not shown), etc.

Network computer environment

[00077] FIG. 2 illustrates a networked computer environment 200 suitable for use with some embodiments of systems and/or methods of the present technology. The networked computing environment 200 includes an electronic device 210 associated with a self-driving vehicle (SDC) 220 or associated with a user (not shown) that can control the SDC 220, a server 235 communicating with the electronic device 210 via a communications network 240 (e.g., the Internet and etc., as will be described in more detail below). Optionally, the networked computing environment 200 may also include a GPS satellite (not shown) transmitting and/or receiving a GPS signal to/from the electronic device 210. It should be understood that the present technology is not limited to GPS and may use location technology other than GPS. It should be noted that the GPS satellite may not be used at all.

[00078] The SDC 220 to which the electronic device 210 is coupled may include any vehicle or vehicle, such as a private or commercial automobile, truck, motorcycle, or the like. The vehicle may be user driven or a driverless vehicle. It should be noted that the specific parameters of SDC 220 are not limiting, these specific parameters include: vehicle manufacturer, vehicle model, vehicle year, vehicle weight, vehicle dimensions, vehicle weight distribution, vehicle surface area, vehicle height, transmission type (eg 2x or 4x), tire type, brake system, fuel system, mileage, vehicle identification number and engine size.

[00079] The implementation of electronic device 210 is not particularly limited, but by way of example, electronic device 210 may be implemented as a vehicle engine control unit, a vehicle central processing unit, a vehicle navigation device (e.g., TomTom™, Garmin™), a tablet, and a personal computer. computer built into SDC 220, etc. Thus, it should be noted that electronic device 210 may or may not be permanently coupled to SDC 220. Additionally or alternatively, electronic device 210 may be implemented in a wireless communications device such as a mobile phone (eg, a smartphone or cordless phone). In certain embodiments, electronic device 210 has a display 270.

[00080] Electronic device 210 may include some or all of the components of the computer system 100 depicted in FIG. 1. In some embodiments, the electronic device 210 is an on-board computer device and includes a processor 110, a solid state drive 120, and a memory 130. In other words, the electronic device 210 includes hardware and/or software and/or firmware, or a combination thereof, for determining trajectory SDC 220 on a given road section, taking into account obstacles on it, as will be described in more detail. below.

Sensory system

[00081] In non-limiting embodiments of the present technology, electronic device 210 includes or has access to a sensor system 230. According to these embodiments, the sensor system 230 may include multiple sensors, allowing for different implementations of the present technology. Examples of multiple sensors include, but are not limited to: cameras, lidar sensors, radar sensors, etc. The sensor system 230 is operatively coupled to the processor 110 to transmit the information thus captured to the processor 110 for processing, as will be described in more detail below herein.

[00082] The touch system 230 may be installed in the inner, upper portion of the windshield SDC 220, but other locations are within the scope of the present technology, including on the rear window, side windows, front hood, roof, front grille, or front bumper of the SDC 220. In some non-limiting embodiments of the present technology, the sensor system 230 may be mounted in a special housing (not shown) mounted on top of the SDC 220.

[00083] In addition, the spatial arrangement of the sensor system 230 may be designed based on its specific technical configuration, housing configuration, weather conditions in the area where the SDC 220 will be used (eg, frequent rain, snow, and other elements), or the like.

[00084] In non-limiting embodiments of the present technology, the sensor system 230 may include a sensor configured to capture an image of the environment 260. In this regard, the sensor system 230 may be a camera or a plurality thereof (not separately depicted).

[00085] How the camera is implemented is not particularly limited. For example, in one specific non-limiting embodiment of the present technology, the camera may be implemented as a mono camera with a resolution sufficient to detect objects at predetermined distances up to approximately 30 m (although cameras with other resolutions and ranges are within the scope of the present technology).

[00086] In some embodiments of the present technology, the camera (or one or more cameras comprising the implementation of the sensor system 230) is configured to capture a predetermined portion of the surrounding area 260 around the SDC 220. In some embodiments of the present technology, the camera is configured to capture an image ( or a series of images) that represent approximately 90 degrees of the surrounding area 260 around the SDC 220 that is along the path of motion of the SDC 220.

[00087] In other embodiments of the present technology, the camera is configured to capture an image (or series of images) that represents approximately 180 degrees of the surrounding area 260 around the SDC 220 that is along the path of motion of the SDC 220. In yet further embodiments of the present technology, the camera is configured with the ability to capture an image (or series of images) that represent approximately 360 degrees of the surrounding area 260 around the SDC 220 that is along the path of movement of the SDC 220 (in other words, the entire surrounding area around the SDC 220).

[00088] In a specific non-limiting example, the camera may be of the type available from FLIR Integrated Imaging Solutions Inc., 12051 Riverside Way, Richmond, British Columbia, V6W 1K7, Canada. It should be clearly understood that the camera may be implemented in any other suitable equipment.

[00089] In non-limiting embodiments of the present technology, sensor system 230 may further comprise a lidar device (not separately illustrated). Lidar stands for optical detection and ranging (LIDAR, Light Detection and Ranging). One of ordinary skill in the art would be expected to understand the functionality of a lidar instrument, but in short, a lidar transmitter (not shown) sends out a laser pulse and light particles (photons) are scattered back into the lidar instrument. The photons that return to the receiver are collected by the telescope and counted as a function of time. Using the speed of light (~3 × 10 ⁸ m/s), processor 110 can then calculate how far the photons have traveled (in a circular path). Photons can be scattered from many different objects surrounding the SDC 220, such as other particles (aerosols or molecules) in the atmosphere, other maps, stationary objects, or potential obstacles in front of the SDC 220.

[00090] In a specific non-limiting example, the lidar device included in the sensor system 230 may be implemented as a lidar-based sensor, which may be of the type available from Velodyne LiDAR, Inc., 5521 Hellyer Avenue, San Jose, CA 95138, United States America. It should be clearly understood that the lidar device can be implemented in any other suitable equipment.

[00091] In some embodiments of the present technology, the lidar device included in the sensor system 230 may be implemented as a plurality of lidar sensors, such as three or any other suitable number.

[00092] In non-limiting embodiments of the present technology, sensor system 230 may further comprise a radio detection and ranging (RADAR) device (not separately illustrated). In short, a radar instrument is a detection tool that uses radio waves to determine the range, angle, and/or speed of objects. A radar instrument includes a transmitter that generates electromagnetic waves, an antenna used to transmit and receive electromagnetic waves, a receiver, and a processor to determine the properties of detected objects.

[00093] In alternative embodiments of the present technology, there may be a separate antenna for receiving waves and a separate antenna for transmitting waves. The processor used to determine the properties of surrounding objects may be processor 110.

[00094] In some embodiments of the present technology, the radar instrument used in the sensor system 230 may include long-range, mid-range, and short-range radar sensors. As a non-limiting example, a long-range radar sensor may be used for adaptive cruise control, automatic emergency braking, and forward collision warning, while a mid- and short-range radar sensor may be used for parking assist, cross-traffic alert, intersection assist and blind side detection.

[00095] In a specific non-limiting example, the radar device contained in the sensor system 230 may be of the type available from Robert Bosch GmbH, Robert-Bosch-Platz 1, 70839 Gerlingen, Germany. It should be clearly understood that the radar instrument may be implemented in any other suitable equipment.

[00096] In some non-limiting embodiments of the present technology, the sensor system 230 may be used by the processor 110 for image calibration. For example, using an image captured by a camera and a lidar point cloud captured by the lidar instrument, processor 110 is configured to identify a predetermined region of the image to correspond to a predetermined region of the lidar point cloud captured by the lidar instrument. In other embodiments of the present technology, the sensor system 230 is calibrated such that, for an image captured by a camera, a lidar point cloud captured by a lidar instrument, and radar data captured by a radar instrument, the processor 110 is configured to identify a given region of the image to correspond to a given areas of lidar point cloud and radar data.

[00097] In non-limiting embodiments of the present technology, SDC 220 further contains or has access to other sensors (not separately shown). Other sensors include one or more of the following: inertial measurement unit (IMU), global navigation satellite system (GNSS) instrument, ground speed radars, ultrasonic sonar sensors, odometry sensors including accelerometers and gyroscopes, mechanical tilt sensors, magnetic compass and other sensors that allow you to work with SDC 220.

[00098] As a non-limiting example, an IMU may be attached to the SDC 220 and contain three gyroscopes and three accelerometers to provide data about the rotational motion and linear motion of the SDC 220, which can be used to calculate the motion and position of the SDC. 220.

Communication network

[00099] In some embodiments of the present technology, the communications network 240 is the Internet. In alternative, non-limiting embodiments, communications network 240 may be implemented as any suitable local area network (LAN), wide area network (WAN), private communications network, or the like. It should be clearly understood that implementations of communication network 240 are shown for illustrative purposes only. How the communication link (not specifically numbered) between the electronic device 210 and the communication network 240 is implemented will depend, in part, on how the electronic device 210 is implemented. By way of example, and not limitation, in those non-limiting embodiments of the present technology, where the electronic device 210 is implemented as a wireless communication device, such as a smartphone or navigation device, the communication link may be implemented as a wireless communication link. Examples of wireless communication links include, but are not limited to, a 3G network link, a 4G network link, a 5G network link, and the like. Communications network 240 may also use a wireless connection to server 235.

Server

[000100] In some embodiments of the present technology, server 235 is implemented as a conventional computer server and may contain some or all of the components of the computer system 100 shown in FIG. 1. In one non-limiting example, server 235 is implemented as a Dell™ PowerEdge™ server running the Microsoft™ Windows Server™ operating system, but may also be implemented in any other suitable hardware, software and/or firmware, or combination thereof. In the depicted non-limiting embodiments of the present technology, the server is a single server. In alternative non-limiting embodiments of the present technology (not shown), the functionality of server 235 may be distributed and implemented by multiple servers.

[000101] In some non-limiting embodiments of the present technology, the processor 110 of the electronic device 210 may communicate with the server 235 to receive one or more updates. Updates may include, but are not limited to, software updates, map updates, route updates, weather updates, etc.

[000102] In some embodiments of the present technology, processor 110 may also be configured to transmit certain operational data, such as routes traveled, traffic data, performance data, and the like, to server 235. Some or all of the data transmitted between SDC 220 and server 235 may be encrypted and/or anonymized.

Storage

[000103] FIG. 2 also depicts storage 250 associated with communication capability with server 235. However, in some embodiments, storage 250 may be associated with communication capability with electronic device 210 and/or may be implemented in electronic device 210 and /or may be associated with the ability to communicate with any other processor of the networked computer environment 200.

[000104] In at least some embodiments, it is contemplated that storage 250 may be used by server 235, electronic device 210, and/or any other processor of networked computing environment 200 as a memory device for storing information. The storage 250 is configured to store information retrieved, determined, and/or generated by the processor 110 of the server 235 and/or the electronic device 210. Generally speaking, the storage 250 may receive data from the processor 110 that was generated by the processor 110 during processing for its temporary storage. and/or persistent storage, and may provide the stored data to the processor 110 for use. It is contemplated that the storage 250 may be divided into multiple distributed storages, for example, to provide a fault-tolerant storage system, without going beyond the scope of the present technology.

[000105] It should be noted that in at least some embodiments of the present technology, storage 250 may be implemented locally on electronic device 210 and/or server 235 (eg, in local memory). However, it is also contemplated that storage 250 may be implemented remotely from electronic device 210 and/or server 235 (eg, in remote memory).

Electronic device

[000106] FIG. 3 depicts a schematic diagram of an electronic device 210 for implementing various non-limiting embodiments of the present technology. Electronic device 210 may contain one or more “modules” to perform a variety of data processing tasks. Some of the non-limiting modules may include a training module 302, a perception module 304, a trajectory generation module 306, a trained machine learning algorithm (MLA) module 308, and a motion planning module 310.

[000107] The illustrated connections between the various modules in FIG. 3 correspond to some non-limiting embodiments of the present technology. Different modules can communicate with each other regardless of the connections shown. It should be noted that electronic device 210 may include additional components (eg, those included in computer system 100) and/or modules, but these are excluded from FIG. 3 for simplicity. It should be noted that each of the non-exhaustive list of modules listed above may be implemented as one or more computer-implemented procedures that are configured to perform data processing and/or data generation as described herein. Alternatively, the module may be a physical electronic device. Optionally, multiple modules (or all modules) may be implemented in a single electronic device 210.

[000108] In at least some embodiments, developers of the present technology have developed methods and devices that can provide solutions in situations where objects may be in the vicinity of the SDC 220 and may pose a risk of collision with the SDC 220 during a maneuver. In addition to providing safe maneuvering, the developed methods and devices can also provide smooth and convenient maneuvering.

[000109] It is intended that "object" may refer to any vehicle (e.g., another SDC, another human-driven vehicle, etc.), a bicycle, a person on the road, a person and pet on the road (in some cases a person holding a pet may be referred to as one entity) or the like. Additionally, if an object is in motion, it may be called a dynamic object; others may be called static objects.

[000110] In some non-limiting embodiments, electronic device 210 may rely on different trained MLAs to predict safe trajectories for SDC 220. In various non-limiting embodiments, electronic device 210 may be configured to learn different MLAs before use. In particular, the training module 302 may be configured to collect training data sets and train various MLAs based on the collected training data sets.

[000111] To collect training data sets and train the MLA, in various non-limiting embodiments, the SDC 220 may operate in either a manual mode or an unmanned (autonomous) mode. FIG. 4 illustrates a high-level functional block diagram of various modules included in training module 302, in accordance with various non-limiting embodiments of the present technology. As illustrated, training module 302 may include a manual mode trajectory collection module 302-2, an unmanned mode trajectory collection module 302-4, a training data set store 302-6, and an MLA training module 302-12. The training module 302-12 may include a ranking MLA 302-14 and a ranking MLA 302-16. It should be noted that training module 302 may include other components and modules, but they are not shown in FIG. 4 for simplicity and illustration.

[000112] In various non-limiting embodiments, ranking MLA 302-14 may be trained to predict a ranking score such that the ranking score may indicate similarity between a corresponding candidate trajectory and a reference trajectory. To collect a set of training data for training the ranking MLA 302-14, in various non-limiting embodiments, the SDC 220 may be operated in a manual mode, that is, the SDC 220 is controlled manually by a human expert 290 (as shown in FIG. 2). In various non-limiting embodiments, a human expert 290 may operate the SDC 220 on a training road segment.

[000113] FIG. 5 depicts a typical training road section 400 in accordance with various non-limiting embodiments of the present technology. As shown, the training road portion 400 may include training road lanes 402, 404, and 406. To collect a set of training data for training the ranking MLA 302-14, in various non-limiting embodiments, a human expert 290 (as shown in FIG. 2) may manually operate the SDC 220 on a road segment 400 for training.

[000114] The manual trajectory collection module 302-2 (as shown in FIG. 4) may be configured to collect the manual training trajectory 408 of the SDC 220 performed by a human expert 290 while the SDC 220 is operating on the training road section 400. The manual training trajectory 408 may include current and future location points, collectively referred to as location points 410, at different times along the training road section 400. As will be discussed in more detail, in various non-limiting embodiments, the manual training path 408 may be used as a reference for training the ranking MLA 302-14, hence, the manual training trajectory 408 may also be referred to as the reference trajectory 408 hereinafter.

[000115] It should be noted that for simplicity, only one manual training path 408 is illustrated in FIG. However, in various non-limiting embodiments, human expert 290 may perform more than one manual training path on different training road sections different from training road section 400. The manual path collection unit 302-2 can collect all such manual learning paths. In addition, the training road section 400 may include other objects (not shown), and the human expert 290 may view such objects when operating the SDC 220.

[000116] At the same time that the human expert 290 operates the SDC 220, the electronic device 210 may be configured to generate predicted training trajectories for the SDC 220. The predicted training trajectories may be generated in the same manner as if the SDC 220 controlled in unmanned mode. However, such predicted training trajectories may not affect the operation of the SDC 220, which is controlled by a human expert 290.

[000117] In some non-limiting embodiments, to generate predicted trajectories for training, the electronic device 210 may be configured to process information received by the sensor system 230 (as shown in FIG. 3) about the environment of the SDC 220. As discussed previously, the SDC sensor system 230 220 may include sensors (e.g., cameras, lidar sensors, radar sensors, etc.) that are configured to obtain information about the environment in real time. The real-time information may include information related to an area in front of, next to, and/or behind the SDC 220. The sensor system 230 may continuously detect and track one or more objects in the vicinity of the SDC 220 to determine parameters regarding driving situations for the objects . Some of the non-limiting parameters include, for example, speed, difference in location of an object from the SDC 220 over time, or the like.

[000118] In some non-limiting embodiments, the sensor system 230 may provide real-time information about the environment of the SDC 220 to the perception module 304 (as shown in FIG. 3). The perception module 304 may be configured to perform at least some data processing with respect to the real-time information and may provide the processed data to other modules of the electronic device 210. The perception module 304 may perform at least some data processing with respect to the real-time information. to determine, for example, the number of objects currently moving near the SDC 220, whether another object is moving relative to the SDC 220 in the same or adjacent lane, whether another object is moving relative to the SDC 220 in the same or opposite direction, which speed or speed difference relative to SDC 220 of this object, information about the steering of the object, etc. It is intended that the above list is not exhaustive, and the perception module 304 can determine, retrieve, or output any suitable information about the environment of the SDC 220 by processing the information in real time.

[000119] In one non-limiting embodiment, perception module 304 may provide processed data to trajectory generation module 306. The trajectory generation module 306 may be configured to generate predicted training trajectories associated with the SDC 220. Each of the predicted training trajectories may include potential future location points of the SDC 220 on the training road section 400 as a function of time. The trajectory generation module 306 may generate predicted training trajectories associated with the SDC 220 based on motion data that is indicative of kinematic characteristics (e.g., speed, acceleration, direction, angle, etc.) of the SDC 220, and based on the site 400 roads for training, which indicates the various restrictions of road sections, such as, for example, traffic rules and the shape of the road.

[000120] Referring to FIG. 5, two of the predicted training trajectories generated by the trajectory generation module 306 may be referred to as predicted training trajectories 412 and 416. Predicted training trajectories 412 and 416 may include current and future location points, collectively referred to as location points 414 and 418, respectively, at different points in time along the training road section 400. It should be noted that the manner in which the predicted training trajectories 412 and 416 are generated by the trajectory generation module 306 should not limit the scope of the present technology.

[000121] FIG. 6 depicts an exemplary timeline 460 including training times 462-1, 462-2, 462-3, and 462-4, in accordance with various non-limiting embodiments of the present technology. In some non-limiting embodiments, the manual trajectory collection module 302-2 (as shown in FIG. 4) may be configured to collect manual training trajectories of the SDC 220 at various training times. At the same time, the manual trajectory collection module 302-2 may be configured to collect predicted trajectories 412 and 416 for training the SDC 220 at different training times.

[000122] As an example, the manual training trajectory collection module 302-2 may be configured to store the manual training trajectory 408 and the predicted training trajectories 412 and 416 as the first training data set 302-8 (as shown in FIG. 4) in the storage 302-6 sets of training data. It should be noted that the first training data set 302-8 may include multiple manual training trajectories 408, also referred to as reference trajectories 408, performed by a human expert 290 on a training road section 400, and corresponding multiple predicted training trajectories, including trajectories 412 and 416 for training, generated by the trajectory generation module 306. In some non-limiting embodiments, the first training data set 302-8 may also include information (eg, number of lanes, width of each lane, curvature of lanes, etc.) about the road segment 400 to be trained.

[000123] In various non-limiting embodiments, the human expert 290 for manual operation of the SDC 220 may be a highly skilled professional driver. In other non-limiting embodiments, human expert 290 may be an ordinary driver trained to operate the SDC 220 in a manner consistent with good driving behavior. The manual training trajectory 408 may offer a smooth and comfortable maneuver compared to the predicted training trajectories 412 and 416. As examples, the acceleration and/or deceleration of the SDC 220 during the manual training path 408 may be less abrupt compared to the predicted training path 412.

[000124] In various non-limiting embodiments, the SDC 220 may be controlled in unmanned mode to generate a second set of training data 302-10 (as shown in FIG. 4) for training the classifier MLA 302-16. In various non-limiting embodiments, the classifying MLA 302-16 can be trained to generate a safety score such that the safety score can indicate whether a corresponding candidate trajectory is safe. It should be noted that the order in which the first training data set 302-8 and the second training data set 302-10 are generated should not limit the scope of the present technology.

[000125] To collect the second training data set 302-10 for training the classifier MLA 302-16, in various non-limiting embodiments, the SDC 220 may be operated in an unmanned mode, that is, the SDC 220 is driven autonomously. However, in various non-limiting embodiments, human expert 290 may still move to SDC 220 at training road section 400 to provide implicit and/or explicit input.

[000126] FIG. 7 shows another view of a training road section 400 in accordance with various non-limiting embodiments of the present technology. To collect a second set of training data 302-10 for training the classifying MLA 302-16, in various non-limiting embodiments, the SDC 220 may be operated in an unmanned mode on the road section 400 for training.

[000127] The trajectory collection module 302-4 in unmanned mode (as shown in FIG. 4) may be configured to collect predicted trajectories 420 and 424 for training SDC 220 generated by the trajectory generation module 306 in a similar manner as discussed previously. Predicted training trajectories 420 and 424 may include current and future location points, collectively referred to as location points 422 and 426, respectively, at different times along the training road section 400. It should be noted that for simplicity, only two predicted training trajectories 420 and 424 are shown in FIG. 7. However, in various non-limiting embodiments, more than two predicted training trajectories may be generated by trajectory generation module 306 at different training road sections different from training road section 400. The trajectory collection module 302-4 in unmanned mode can collect all such predicted trajectories for training.

[000128] In addition, the training road section 400 may include an object 430 traveling along the training road lane 404. It should be understood that the training road portion 400 may include other objects but are not shown in FIG. 7 for simplicity.

[000129] FIG. 8 depicts a typical timeline 470 including training time points 472-1, 472-2, 472-3, and 472-4, in accordance with various non-limiting embodiments of the present technology. In some non-limiting embodiments, the unmanned mode trajectory collection module 302-4 (as shown in FIG. 4) may be configured to collect predicted trajectories 420 and 424 for training SDC 220 at various training time points 472-1 and 472-2, respectively. . In various non-limiting embodiments, training times 472-1, 472-2, 472-3, and 472-4 may correspond to “ticks” (time stamps) that represent periodic times during which trajectory generation module 306 may be executed with the ability to generate new/updated predicted trajectories for controlling the SDC 220 and collecting predicted trajectories 420 and 424 for training the SDC 220 generated by the trajectory generation module 306. In some non-limiting embodiments, the duration between two training times, such as training times 472-1 and 472-2, may be predetermined.

[000130] As discussed previously, even though the SDC 220 may be controlled in unmanned mode, the human expert 290 may still be driving the SDC 220 on the training road section 400 during times 472-1, 472-2, 472 -3 and 472-4 training times to provide implicit and/or explicit input 478. In some non-limiting embodiments, unmanned mode trajectory acquisition module 302-4 may be configured to provide a label for predicted trajectories 420 and 424 for training. The label may indicate at least one of the explicit and implicit inputs of the human expert 290.

[000131] In some non-limiting embodiments, implicit input 478 may be referred to as the indirect agreement of a human expert 290 on a predicted trajectory for training (e.g., predicted trajectories 420 or 424 for training). In other words, if human expert 290 does not provide any input while SDC 220 follows a particular predicted training path (e.g., predicted training paths 420 or 424), this may be referred to as implicit input 478. In one example, if The SDC 220 follows the predicted trajectory 420 for training and the human expert 290 does not provide any input and allows the SDC 220 to move straight ahead, the input may be referred to as implicit input 478. In another example, if the SDC 220 follows the predicted trajectory 424 for training and the human expert The expert 290 does not provide any data and allows the SDC 220 to change the training road lane from the training road lane 402 to the training road lane 404, the input may be called implicit input 478. Thus, it can be said that no intervention such as the lack of participation or implied consent of the human expert 290 while the SDC 220 is performing either a lane change maneuver or a lane keeping maneuver can be considered as implicit input 478 for that particular maneuver.

[000132] In some non-limiting embodiments, explicit input 478 may be referred to as disagreement by the human expert 290 with the predicted trajectory for training (eg, predicted trajectories 420 or 424 for training). In other words, if human expert 290 provides input while SDC 220 follows a particular predicted training trajectory (e.g., predicted training trajectories 420 or 424), this may be referred to as explicit input 478. Input from human expert 290 may be in the form of direct action affecting the control of the SDC 220 in unmanned mode. In one example, if SDC 220 follows a predicted trajectory 420 for training and human expert 290 provides input (such as acceleration, deceleration, braking, steering, etc.) resulting in a change in SDC 220's trajectory, this may be referred to as explicit input 478. In another example, if the SDC 220 follows the predicted training trajectory 424 and the human expert 290 provides input and avoids the SDC 220 changing the training road lane from the training road lane 402 to the training road lane 404, input may be referred to as explicit input 478. Thus, it may be said that an intervention, such as the intervention or dissent of a human expert 290 while SDC 220 performs either a lane change maneuver or a lane keeping maneuver, may be referred to as explicit input 478 for this particular maneuver.

[000133] It is contemplated that in the event that a human expert 290 must take over control of the SDC 220, this may be referred to as a situation where the predicted training trajectory 420 or 424 may not be safe to follow. To this end, in some embodiments, if the input associated with a particular predicted training trajectory (eg, predicted training trajectories 420 or 424) is explicit, the trajectory may be marked as an unsafe trajectory. On the other hand, if the input associated with a particular predicted training trajectory (eg, predicted training trajectories 420 or 424) is implicit, the trajectory may be marked as a safe trajectory.

[000134] In some non-limiting embodiments, the unmanned mode trajectory collection module 302-4 may be configured to store predicted training trajectories 420 and 424 as a second training data set 302-10 (as shown in FIG. 4) in storage 302 -6 training data sets. It should be noted that the second training data set 302-10 may include multiple predicted training trajectories labeled as safe or unsafe.

[000135] FIG. 9 depicts training of the ranking MLA 302-14 using the first training data set 302-8 as input in accordance with various non-limiting embodiments of the present technology. To train the ranking MLA 302-14, in some non-limiting embodiments, the MLA training module 302-12 (as shown in FIG. 4) may be configured to obtain the first training data set 302-8 from the training data set storage 302-6. As discussed previously, the first training data set 302-8 may include manual training trajectories 408 and predicted training trajectories 412 and 416.

[000136] Based on the predicted training trajectories 412 and 416 and the manual training trajectories 408, the MLA training module 302-12 may be configured to train a ranking MLA 302-14 to predict a ranking score associated with a candidate trajectory. In some non-limiting embodiments, the ranking score may indicate the similarity between the predicted training trajectories 412 and 416 and the manual training trajectories 408. In other words, ranking MLA 302-14 may be configured to compare predicted training trajectories 412 and 416 with manual training trajectories 408 and may generate a ranking estimate. The ranking score may indicate how similar the predicted training trajectories 412 and 416 are to the manual training trajectory 408 . A more similar trajectory may be ranked higher than a trajectory that is less similar to manual training trajectories 408 .

[000137] In various non-limiting embodiments, the ranking MLA 302-14 can be trained to predict a ranking score without considering the safety of the corresponding training trajectory. The ranking MLA 302-14 may provide a ranking score 480 corresponding to the predicted training trajectory 412 and a ranking score 482 corresponding to the predicted training trajectory 416. In the event that the predicted training trajectory 412 is more similar to the manual training trajectories 408 than the predicted training trajectory 418, the ranking score 480 may be higher than the ranking score 482, and vice versa.

[000138] In some non-limiting embodiments, the ranking MLA 302-14 may be configured to calculate a distance indicating how similar the predicted training trajectories 412 and 416 are to the manual training trajectory 408. Based on distance, ranking MLA 302-14 may be configured to rank predicted trajectories 412 and 416 for training. In some non-limiting embodiments, when training the ranking MLA 302-14, a distance indicating the similarity of the predicted training trajectories 412 and 416 relative to the manual training trajectories 408 may be provided as feedback to the ranking MLA 302-14 for supervised training by the ranking MLA 302-14.

[000139] In some non-limiting embodiments, the ranking MLA 302-14 may be a pairwise ranking algorithm and/or a list ranking algorithm trained in a supervised learning method. However, any other suitable ranking algorithm may be used without departing from the principles of this technology.

[000140] FIG. 10 depicts training a classifier MLA 302-16 using a second set of training data 302-10 as input in accordance with various non-limiting embodiments of the present technology. To train the classifying MLA 302-16, in some non-limiting embodiments, the MLA training module 302-12 (as shown in FIG. 4) may be configured to retrieve a second training data set 302-10 from the training data set storage 302-6. As discussed previously, the second training data set 302-10 may include predicted trajectories 420 and 424 for training. Predicted trajectories 420 and 424 for training may include labels indicating at least one of the implicit and explicit inputs 478 of the human expert 290.

[000141] Based on the predicted trajectories 420 and 424 for training, the MLA training module 302-12 may be configured to train the classifying MLA 302-16 to generate a safety assessment and, based on the safety assessment, classify whether the corresponding potential trajectory is safe. In some non-limiting embodiments, the classifier MLA 302-16 may also be provided with implicit and explicit input 478 from a human expert 290 to train the classifier MLA 302-16.

[000142] Referring to FIG. 10, the classifying MLA 302-16 may provide a safety score 490 corresponding to the predicted training trajectory 420 and a safety score 492 corresponding to the predicted training trajectory 424. The safety score 490 and 492 may indicate whether the predicted trajectories 420 and 424 are safe for training.

[000143] In some non-limiting embodiments, the classifying MLA 302-16 may be a decision tree algorithm trained in a supervised learning manner. However, any other suitable classification algorithm may be used without departing from the principles of the present technology.

[000144] After the ranking MLA 302-14 and the classifying MLA 302-16 are trained by the MLA training module 302-12, the SDC 220 can be driven on a road section other than the training road section 400 with actual traffic. The SDC 220 can rely on the trained ranking MLA 302-14 and the trained classifying MLA 302-16 while driving along a road section.

[000145] FIG. 11 illustrates a use case where the SDC 220 is controlled on a road segment 500, in accordance with various non-limiting embodiments of the present technology. As shown, road section 500 may include traffic lanes 502-1, 502-2, and 502-3. Initially, it can be assumed that the SDC 220 is moving in lane 502-2. It is also assumed that object 504 is moving in traffic lane 502-2. It should be noted that the above road section 500 and use scenario are for illustration purposes only. In various non-limiting embodiments, road section 500 may include any number of lanes with any number of objects.

[000146] Referring to FIG. 3, in various non-limiting embodiments, at a given time while the SDC 220 is operating on a road segment 500, the sensor system 230 may be configured to obtain real-time information about the environment of the SDC 220. The sensor system 230 may provide real-time information about the environment of the SDC 220 to the perception module 304. As discussed previously, perception module 304 may be configured to perform at least some data processing with real-time information. The perception module 304 may be configured to provide the processed information in real time to the trajectory generation module 306. Path generation module 306 may generate potential paths associated with SDC 220 based on motion data that is indicative of kinematic characteristics (e.g., speed, acceleration, direction, angle, etc.) of SDC 220, and based on road segment 500, which indicates the various restrictions of the road sections, such as, for example, traffic rules and the shape of the road.

[000147] Referring to FIG. 11, candidate trajectories generated by the trajectory generating unit 306 can be represented as candidate trajectories 506-1, 506-2, and 506-3. It should be noted that for simplicity, only three trajectories are shown. However, in various non-limiting embodiments, any number of trajectories may be generated by trajectory generation module 306.

[000148] The trajectory generation module 306 may be configured to route candidate trajectories 506-1, 506-2, and 506-3 to the trained MLA module 308 (as shown in FIG. 3). In some non-limiting embodiments, training module 302 may be configured to provide a pre-trained ranking MLA 302-12 and classifying MLA 302-16 to the trained MLA module 308. For simplicity and clarity, the previously trained MLA 302-12 may be referred to as the trained ranking MLA 308-1, and the previously trained classifying MLA 302-16 may be referred to as the trained ranking MLA 308-2.

[000149] FIG. 12 illustrates a trained ranking MLA 308-1 trained to rank candidate trajectories 506-1, 506-2, and 506-3, in accordance with various non-limiting embodiments of the present technology. For simplicity, only candidate trajectories 506-1 and 506-2 are shown as input to the trained ranking MLA 308-1. However, in various non-limiting embodiments, any number of candidate trajectories may be provided as input to the trained ranking MLA 308-1. The trained ranking MLA 308-1 may be configured to rank candidate trajectories 506-1 and 506-2. To do this, the trained ranking MLA 308-1 may be trained to predict ranking scores for candidate trajectories 506-1 and 506-2.

[000150] According to the candidate trajectory 506-1, the trained ranking MLA 308-1 can predict the ranking score 510-1. Likewise, for candidate trajectory 506-2, the trained ranking MLA 308-1 can predict the ranking score 510-2.

[000151] In various non-limiting embodiments, the trained ranking MLA 302-14 may predict ranking scores 510-1 and 510-2 indicating similarity between candidate trajectories 506-1 and 506-2 and a reference trajectory (not illustrated). In some non-limiting embodiments, the ranking scores may indicate how similar potential trajectories 506-1 and 506-2 are to a given reference trajectory that would be performed by a human expert 290 on road segment 500. It should be noted that the ranking scores 510-1 and 510-2 may be predicted without regard to the safety of the corresponding candidate trajectories 506-1 and 506-2. In other words, potential trajectories 506-1 and 506-2 may or may not be safe for SDC 220 to maneuver.

[000152] In FIG. 13 shows a trained classifier MLA 308-2 trained to classify candidate trajectories 506-1, 506-2, and 506-3 in accordance with various non-limiting embodiments of the present technology. For simplicity, only candidate trajectories 506-1 and 506-2 are shown as input to the trained classifier MLA 308-2. However, in various non-limiting embodiments, any number of candidate trajectories may be provided as input to the trained classifier MLA 308-2. The trained classifier MLA 308-2 may be configured to classify candidate trajectories 506-1 and 506-2. To achieve this, the trained classifier MLA 308-2 may be trained to generate safety scores for candidate trajectories 506-1 and 506-2.

[000153] According to the candidate trajectory 506-1, the trained classification MLA 308-2 may generate a safety score 512-1. Similarly, for candidate trajectory 506-2, the trained classifier MLA 308-2 may generate a safety score 512-2.

[000154] It should be noted that the trained ranking MLA 308-1 and the trained classifying MLA 308-2 may have different computer architectures having different functionality. As an example, the trained ranking MLA 308-1 may perform ranking of candidate trajectories (such as potential trajectories 506-1 and 506-2), while the trained classifying MLA 308-2 may perform classification of candidate trajectories (such as potential trajectories 506 -1 and 506-2). In addition, the trained ranking MLA 308-1 and the trained classifying MLA 308-2 can be independently trained using different training data sets, such as the first training data set 302-8 and the second training data set 302-10, respectively.

[000155] In various non-limiting embodiments, the trained ranking MLA 308-1 and the trained classifying MLA 308-2 may be configured to operate in parallel. To this end, the classification of candidate trajectories 506-1 and 506-2 may be performed in parallel with the ranking of candidate trajectories 506-1 and 506-2. Additionally, in various non-limiting embodiments, the classification of candidate trajectories 506-1 and 506-2 may be performed independently of the ranking of candidate trajectories 506-1 and 506-2.

[000156] FIG. 14 shows a trajectory identification module 310-2 implemented as a submodule within a motion planning module 310, in accordance with various non-limiting embodiments of the present technology. The trained MLA module 308 may be configured to provide candidate trajectories 506-1 and 506-2 along with ranking scores 510-2 and 510-2 and security scores 512-1 and 512-2 to trajectory identification module 310-2. It should be noted that for simplicity, only the candidate trajectories 506-1 and 506-2 are shown as input to the trajectory identification module 310-2. However, in various non-limiting embodiments, any number of candidate trajectories may be provided as input to the trajectory identification module 310-2. The trajectory identification module 310-2 may be configured to identify the best trajectory 520 among the candidate trajectories 506-1 and 506-2 and the safest trajectory 522 among the candidate trajectories 506-1 and 506-2 based on a combination of the respective scores 510-2 and 510 -2 rankings and ratings 512-1 and 512-2 security.

[000157] In some non-limiting embodiments, the best trajectory 520 may be the trajectory with the highest ranking score among the ranking scores 510-1 and 510-2. In addition, the safest trajectory 522 may be the trajectory with the highest safety score among the safety scores 512-1 and 512-2, and which is also labeled as safe (using the classification model). For example, if the first and second ranked trajectories are not marked as safe according to their ranking, while the third ranked trajectory is marked as safe, the trajectory identification module 310-2 may select the third ranked trajectory as the safest trajectory 522.

[000158] In some non-limiting embodiments, motion planning module 310 may cause SDC 220 to operate in accordance with the safest trajectory 522. In some non-limiting embodiments, if the safest trajectory 522 and the best trajectory 520 are not the same candidate trajectory, the module Motion planning 310 may direct SDC 220 to take corrective action by operating in accordance with the safest trajectory 522. In some non-limiting embodiments, the corrective action may cause the best trajectory 520 to become the safest trajectory 522 at a future point in time.

[000159] In some non-limiting embodiments, the corrective action may be to turn on the visual indicator 508 (as shown in FIG. 16). The visual indicator 508 may indicate the intention of the SDC 220 to perform a lane change maneuver with respect to the object 504. Thus, a change in the kinematic characteristics of the object 504 can be expected. To this end, the trajectory identification module 310-2 may again identify the best trajectory 520 and the safest trajectory 522 based on changed conditions on section 500 of the road.

[000160] Additionally, in some non-limiting embodiments, based on the best trajectory 520 and the safest trajectory 522, the motion planning module 310 may be configured to cause the SDC 220 to perform the first or second action. In some non-limiting embodiments, the first action is invoked by motion planning module 310 if the best trajectory 520 and the safest trajectory 522 are the same candidate trajectory (eg, candidate trajectory 506-1 or 506-2). In some non-limiting embodiments, a second action is invoked by motion planning module 310 if the best trajectory 520 and the safest trajectory 522 are not the same candidate trajectory (eg, candidate trajectory 506-1 or 506-2).

[000161] FIG. 15 illustrates a schematic diagram of a state transition engine 310-4 implemented in motion planning module 310, in accordance with various non-limiting embodiments of the present technology. The state transition mechanism 310-4 may include various states, such as a lane keeping maneuver state 602, a lane change maneuver preparation state 604, and a lane change maneuver state 606.

[000162] Various states, such as lane keep maneuver state 602, lane change maneuver preparation state 604, and lane change maneuver state 606, may correspond to different driving conditions of the SDC 220. As an example, if the current state is state 602 lane keeping maneuver, SDC 220 may continue driving in the current lane (eg, lane 502-2). In another example, if the current state is a lane change maneuver preparation state 604, motion planning module 310 may make various on-the-fly adjustments to the current kinematics of SDC 220. Such adjustments may include, but are not limited to, applying braking, reducing speed, reducing acceleration, and etc. In addition, motion planning module 310 may cause SDC 220 to flash a visual indicator 508 indicating an intent to perform a potential lane change maneuver.

[000163] In some non-limiting embodiments, the input to the state transition engine 310-4 may be the best trajectory 520, the safest trajectory 522, and candidate trajectories 506-1 and 506-2. It should be noted that for simplicity, only two potential paths 506-1 and 506-2 have been considered, however, in various non-limiting embodiments, any number of paths may be provided as input to state transition engine 310-4.

[000164] In one non-limiting example, candidate path 506-1 may comprise a potential lane change maneuver, such as a potential lane change maneuver from lane 502-2 to lane 502-3 (as shown in FIG. 11). In addition, the potential path 506-2 may include a potential lane-keeping maneuver, such as a potential lane-keeping maneuver in the lane 502-2 in which the SDC 220 is currently driving.

[000165] Initially, it can be assumed that the state transition mechanism 310-4 is in the lane keeping maneuver state 602, and the SDC 220 can move along the traffic lane 502-2, and the best trajectory 520, the safest trajectory 522, and potential trajectories 506- 1 and 506-2 may be provided as input to state transition engine 310-4.

[000166] In some non-limiting embodiments, if both the best trajectory 520 and the safest trajectory 522 are candidate trajectory 506-1, the state transition mechanism 310-4 may remain in the lane keeping maneuver state 602. To this end, the first action called by the motion planning module 310 may attempt to perform a potential lane keeping maneuver based on the potential trajectory 506-1. In other words, SDC 220 can continue driving in the current lane 502-2.

[000167] In some non-limiting embodiments, if both the best trajectory 520 and the safest trajectory 522 are candidate trajectory 506-2, the state transition mechanism 310-4 may enter the lane change maneuver state 606. To this end, the first action called by the motion planning module 310 may attempt to perform a potential lane change maneuver based on the potential trajectory 506-2. In some non-limiting embodiments, after performing a potential lane change maneuver, the state transition mechanism 310-4 may return to the lane keeping maneuver state 602.

[000168] In some non-limiting embodiments, if the best trajectory 520 is a potential trajectory 506-2 and the safest trajectory 522 is a potential trajectory 506-1, the state transition mechanism 310-4 may enter a lane change maneuver preparation state 604 . To this end, the second action invoked by the motion planning module 310 may include a visual indicator 508 of the SDC 220 to indicate an intent to perform a potential lane change maneuver while executing the potential lane keeping maneuver to another predetermined time point after the given time point.

[000169] FIG. 16 illustrates a typical scenario of the SDC 220 being controlled on a road segment 500 when the state transition mechanism 310-4 is in a lane change maneuver preparation state 604, in accordance with various non-limiting embodiments of the present technology. Since in this scenario the best trajectory 520 may differ from the safest trajectory 522, the electronic device 210 may trigger the visual indicator 508 while the SDC 220 maintains its lane. In other words, the SDC 220 may continue to move in the current lane 502-2 while the traffic planning module 310 may include a visual indicator 508 of the SDC 220 to indicate an intent to perform a potential lane change maneuver.

[000170] Without wishing to be bound by any particular theory, the developers of the present technology have realized that running the visual indicator 508 in a scenario where the best trajectory 520 is a potential trajectory 506-2 and the safest trajectory 522 is a potential trajectory 506-1 can influence the operator of another vehicle (such as object 504) to potentially "give way" or provide SDC 220 more time and/or space to perform a lane change maneuver. As will be described below, on the next tick, for example, when the electronic device 210 is configured to determine a new best trajectory 524 and a new safest trajectory 526, the ability to activate the visual indicator 508 on the current tick may cause the object 504 to behave as such. in a manner that will change the classification of the potential trajectory 506-2 at the next tick, that is, at the next tick the classification model may determine, based on the behavior of the object 504 under the influence, that the potential trajectory 506-2 is now safe. Therefore, at the next tick, potential trajectory 506-2 may be either the best or safest trajectory.

[000171] It is expected that one skilled in the art will understand that any type of visual indicator 508 may be included in the SDC 220. One non-limiting example of a visual indicator 508 may include light-emitting diodes (LEDs) mounted as a visual indicator 508 on the left and the right side of the outer body of the SDC 220. If the intention is to change lanes to the driving lane 502-3 to the left of the SDC 220, the traffic planning module 310 may turn on the visual indicator 508 mounted on the left side of the SDC 220. Additionally, if intent is to change lanes to the traffic lane 502-3 to the right of the SDC 220, the traffic planning module 310 may turn on the visual indicator 508 mounted on the right side of the SDC 220.

[000172] Additionally, in various non-limiting embodiments, traffic planning module 310 may be configured to make various operational adjustments to the current kinematics of SDC 220 to adapt to current traffic conditions on road section 500. As shown in FIG. 16, the previous position SDC 220 is represented as the previous position 220-1. In addition, the previous position of the object 504 is represented as the previous position 504-1. It should be noted that the previous position 220-1 and the previous position 504-1 may be associated with a given point in time.

[000173] In one non-limiting example, the current position of SDC 220 at another point in time may be represented as current position 220-2. Regarding the current position of object 504, two different non-limiting examples were considered. In the first example, object 504 may be slightly far from SDC 220, and the current position of object 504 may be represented as current position 504-2. In a second example, object 504 may be adjacent to SDC 220, and the current position of object 504 may be represented as current position 504-3.

[000174] As noted previously, in the event that the best trajectory 520 and the safest trajectory 522 are different trajectories, the SDC 220 may perform a potential lane keeping maneuver and may continue to travel in the travel lane 502-2 until another time. In addition, at another point in time, object 504 may be at the current position 504-2 or 504-3 depending on its kinematic characteristics.

[000175] At another given point in time after a given point in time, the motion planning module 310 may be configured to communicate with the trained MLA module 308 to compute new ranking scores and new safety scores for candidate trajectories 506-1 and 506-2. In turn, the trained ranking MLA 308-1 (as shown in FIG. 12) may be configured to predict new ranking scores 510-3 and 510-4 associated with candidate trajectories 506-1 and 506-2. In addition, the trained classifying MLA 308-2 (as shown in FIG. 13) may be configured to generate new safety scores 512-3 and 512-4 associated with candidate trajectories 506-1 and 506-2. The trained ranking MLA 308-1 and the trained classifying MLA 308-2 may provide new ranking scores 510-3 and 510-4 and new safety scores 512-3 and 512-4 to motion planning module 310.

[000176] In some non-limiting embodiments, motion planning module 310 may communicate with trajectory identification module 310-2 to identify a new best trajectory 524 among candidate trajectories 506-1 and 506-2 and a new safest trajectory 526 among potential trajectories 506- 1 and 506-2 based on a combination of the respective new ranking scores 510-3 and 510-4 and the new safety scores 512-3 and 512-4. Referring to FIG. 14, the path identification module 310-2 may be configured to identify a new best path 524 and a new safest path 526.

[000177] Referring to FIG. 15, in some non-limiting embodiments, where if the state transition engine 310-4 is in the lane change maneuver preparation state 604, and if both the new best trajectory 524 and the new safest trajectory 526 are the potential trajectory 506-2, the state transition mechanism 310-4 may transition to the lane change maneuver state 606. In this case, the motion planning module 310 may direct the SDC 220 to perform a potential lane change maneuver based on the potential trajectory 506-2. In some non-limiting embodiments, after performing a potential lane change maneuver, the state transition mechanism 310-4 may return to the lane keeping maneuver state 602.

[000178] In some non-limiting embodiments, if the state transition engine 310-4 is in the lane change maneuver preparation state 604, and if both the new best path 524 and the new safest path 526 are potential paths 506-1, the engine State transition 310-4 may transition to lane keeping maneuver state 602. In this case, the motion planning module 310 may direct the SDC 220 to perform a potential lane keeping maneuver based on the potential trajectory 506-1.

[000179] In some non-limiting embodiments, if the state transition mechanism 310-4 is in the lane keeping maneuver state 602 and if the best trajectory 520 is a potential trajectory 506-1 and the safest trajectory 522 is a potential trajectory 506-2, the mechanism State transition 310-4 may transition to lane change maneuver state 606. In this case, the motion planning module 310 may direct the SDC 220 to perform a potential lane change maneuver based on the potential trajectory 506-2. In some non-limiting embodiments, after performing a potential lane change maneuver, the state transition mechanism 310-4 may return to the lane keeping maneuver state 602.

Computer-implemented method

[000180] FIG. 17 shows a flow diagram of a method 700 for controlling SDC 220 in accordance with various non-limiting embodiments of the present technology. In some non-limiting embodiments of the present technology, method 700 may be implemented by electronic device 210 communicatively coupled to SDC 220. Method 700 begins at block 702.

Step 702: Electronically generating a first potential path and a second potential path for the SDC on the road section, respectively.

[000181] At step 702, the electronic device 210 generates a first candidate path and a second potential path for SDC 220 on road section 500, respectively. As discussed previously, trajectory generating module 306 may be configured to generate candidate trajectories 506-1 and 506-2.

Step 704: Electronically ranking using a machine learning ranking algorithm (MLA) the first candidate trajectory and the second candidate trajectory, wherein the ranking MLA is trained to predict a similarity ranking score between the corresponding candidate trajectory and the reference trajectory.

[000182] The method 700 proceeds to step 704 where, using the ranking MLA, the electronic device 210 ranks the first candidate trajectory and the second candidate trajectory, wherein the ranking MLA is trained to predict a ranking score indicating similarity between the corresponding candidate trajectory and the reference trajectory. As discussed previously, the trained ranking MLA 308-1 can rank candidate trajectories 506-1 and 506-2. To do this, the trained ranking MLA 308-1 can predict ranking scores 510-1 and 510-2 associated with candidate trajectories 506-1 and 506-2, respectively. The ranking scores 510-1 and 510-2 may indicate similarity between the respective candidate trajectories 506-1 and 506-2 and the reference trajectory.

[000183] In at least some embodiments of the present technology, the electronic device 210 may be configured to rank a plurality of potential trajectories based on their similarity to a reference trajectory that a human operator could perform on a corresponding road segment. It is assumed that the ranking of multiple potential trajectories can be performed in pairs and/or as a list. In addition, we can say that the ranking of a set of trajectories is performed taking into account the similarity of the corresponding one of the set of trajectories with the reference trajectory for a given road section.

[000184] In at least some embodiments, it can also be said that the ranking of multiple trajectories can be performed without taking into account other objects on the road segment. In other words, the purpose of ranking potential trajectories may be to assess how close or similar the corresponding potential trajectories are to a reference for a given road segment, but determining how safe the corresponding potential trajectory is from the point of view of other objects on the road segment may be performed in a separate step method 700 and, in particular, using a separate MLA. For example, electronic device 210 may determine how safe a given potential trajectory is using the classifying MLA at step 706.

Step 706: Classify electronically using a classifying MLA of the first candidate trajectory and the second candidate trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe.

[000185] Method 700 proceeds to step 706 where, using the classifying MLA, the electronic device 210 classifies the first candidate trajectory and the second potential trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe. As noted previously, the trained classifier MLA 308-2 can classify candidate trajectories 506-1 and 506-2. To do this, the trained classifying MLA 308-2 can generate safety scores 512-1 and 512-2 associated with candidate trajectories 506-1 and 506-2, respectively. The safety assessments 512-1 and 512-2 may indicate whether the respective potential trajectories 506-1 and 506-2 could be safe.

Step 708: Electronically identifying the best trajectory among the first candidate trajectory and the second candidate trajectory and the safest trajectory among the first candidate trajectory and the second candidate trajectory based on a combination of the respective ranking scores and the safety scores.

[000186] The method 700 proceeds to step 708, where the electronic device 210 identifies the best trajectory among the first candidate trajectory and the second candidate trajectory and the safest trajectory among the first candidate trajectory and the second trajectory based on a combination of the respective ranking scores and the safety scores. As noted previously, trajectory identification module 310-2 may identify the best trajectory 520 among candidate trajectories 506-1 and 506-2 and the safest trajectory 522 among potential trajectories 506-1 and 506-2 based on a combination of the respective scores 510-1 and 510 -2 rankings and ratings 512-1 and 512-2 security.

Step 710: Causing the SDC electronic device to perform the first action if the best trajectory and the safest trajectory are the same candidate trajectory.

[000187] The method 700 proceeds to step 710, where the electronic device 210 causes the SDC 220 to perform a first action if the best trajectory 520 and the safest trajectory 522 are the same candidate trajectory. As noted previously, the first action is invoked by motion planning module 310 if the best trajectory 520 and the safest trajectory 522 are the same candidate trajectory (eg, candidate trajectory 506-1 or 506-2).

[000188] At certain non-limiting steps, if both the best trajectory 520 and the safest trajectory 522 are a potential trajectory 506-2 (the potential trajectory 506-2 may include a potential lane change maneuver), the first action called by planning module 310 traffic may attempt to perform a potential lane change maneuver based on the potential trajectory 506-2.

[000189] At certain non-limiting steps, if both the best trajectory 520 and the safest trajectory 522 are a potential trajectory 506-1 (the potential trajectory 506-1 may include a potential lane keeping maneuver), the first action caused by the traffic planning module 310 , may attempt to perform a potential lane keeping maneuver based on the potential trajectory 506-1.

Step 712: Causing the SDC electronic device to perform a second action if the best trajectory and the safest trajectory are not the same candidate trajectory.

[000190] The method 700 proceeds to step 710, where the electronic device 210 causes the SDC 220 to perform a second action if the best trajectory 520 and the safest trajectory 522 are not the same candidate trajectory. As noted previously, a second action is called by motion planning module 310 if the best trajectory 520 and the safest trajectory 522 are not the same candidate trajectory (eg, candidate trajectory 506-1 or 506-2).

[000191] At a certain non-limiting step, if the best trajectory 520 is a potential trajectory 506-2 (potential trajectory 506-2 may include a potential lane change maneuver) and the safest trajectory 522 is a potential trajectory 506-1 (potential trajectory 506-1 may contain a potential lane keeping maneuver), traffic planning module 310 may trigger a second action. The second action caused by the motion planning module 310 may include a visual indicator 508 of the SDC 220 to indicate an intent to perform a potential lane change maneuver while executing the potential lane keeping maneuver to another predetermined time point after the given time point.

[000192] Thus, a safe lane change decision can be made using the electronic device 210 and method 700, which significantly reduces the likelihood of collisions and allows the SDC 220 to perform a lane change safely and smoothly.

[000193] Modifications and improvements to the above-described implementations of the present technology may become apparent to those skilled in the art. The foregoing description is intended to be exemplary and not limiting. It is therefore intended that the scope of the present technology be limited only by the scope of the appended claims.

[000194] While the above-described implementations have been described and shown with reference to specific steps performed in a particular order, it should be understood that some of these steps may be combined, subdivided, or rearranged without departing from the principles of the present technology. Accordingly, the order and grouping of the steps mentioned are not limitations of the present technology.

Claims (93)

1. A method for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's surroundings, wherein the SDC is coupled to an electronic device, the method including the steps of: at a given time during SDC operation on a road section: generating, using an electronic device, a first potential trajectory and a second potential trajectory for the SDC on the road section, respectively; ranking, using an electronic device using a machine learning ranking algorithm (MLA), the first potential trajectory and the second potential trajectory, wherein the ranking MLA is trained to predict a ranking score indicating the similarity between the corresponding candidate trajectory and the reference trajectory; classifying by an electronic device using a classifying MLA the first potential trajectory and the second potential trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; electronically identifying the best trajectory among the first potential trajectory and the second potential trajectory and the safest trajectory among the first potential trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; directing the SDC electronic device to perform the first action if the best trajectory and the safest trajectory are the same potential trajectory; And direct the electronic SDC to take a second action if the best path and the safest path are not the same potential path. 2. The method of claim 1, wherein the ranking MLA is a pairwise ranking algorithm that is trained on the first set of training data, wherein the first set of training data contains a reference trajectory representing a training trajectory manually executed by a human expert driving in the SDC on the training road section at the training time; a first predicted training trajectory and a second predicted training trajectory generated by the electronic device for the SDC for the training road section at the training time; the ranking MLA is trained on the first training data set to rank the first predicted training trajectory and the second predicted training trajectory for the training road segment based on corresponding similarity to the reference trajectory. 3. The method of claim 1, wherein the classifying MLA is trained based on the second set of training data, wherein the second set of training data contains a training trajectory generated by the electronic SDC device on the training road section at the training time; a label indicating at least one of the implicit and explicit inputs of a human expert driving in the SDC on the training road section during the training time point. 4. The method of claim 1, wherein the classifying MLA is a decision tree-based algorithm. 5. The method of claim 1, wherein the road section has two lanes and the first potential path comprises a potential lane change maneuver and the second potential path comprises a potential lane keeping maneuver and wherein: if both the best trajectory and the safest trajectory are the first potential trajectory, the first action is to attempt a potential lane change maneuver based on the first potential trajectory; if both the best trajectory and the safest trajectory are the second potential trajectory, the first action is to attempt a potential lane keeping maneuver based on the second potential trajectory. 6. The method of claim 1, wherein the road section has two lanes and the first potential path comprises a potential lane change maneuver and the second potential path comprises a potential lane keeping maneuver and wherein, if the best trajectory is the first potential trajectory and the safest trajectory is the second potential trajectory, the second action is to turn on the visual SDC indicator to indicate the intent to perform the potential lane change maneuver while executing the potential lane keeping maneuver until another predetermined time after the predetermined time. 7. The method of claim 1, wherein the road section has two lanes of traffic and the first potential trajectory comprises a potential lane change maneuver and the second potential trajectory comprises a potential lane keeping maneuver and wherein, if the best trajectory is the second potential trajectory and the safest trajectory is the first potential trajectory, the second action is to attempt a potential lane change maneuver based on the first potential trajectory. 8. The method according to claim 6, wherein the method additionally contains steps in which at another given point in time after a given point in time: electronically identifying a new best trajectory among the first and second potential trajectories and a new safest trajectory among the first potential trajectory and the second potential trajectory based on a combination of the respective new ranking scores and the new safety scores; in response to the new best trajectory and the new safest trajectory, which are both the first potential trajectory, direct the electronic SDC device to perform a potential lane change maneuver based on the first potential trajectory. 9. A method for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's surroundings, wherein the SDC is coupled to an electronic device, the method including the steps of: at a given time during SDC operation on a road section: ranking by means of an electronic device using a ranking MLA the first potential trajectory and the second potential trajectory, wherein the ranking MLA is trained to predict a ranking score indicating the similarity between the corresponding candidate trajectory and the reference trajectory, without considering the safety of the corresponding candidate trajectory; classifying by an electronic device using a classifying MLA the first potential trajectory and the second potential trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; causing the electronic SDC device to perform an action based on a combination of the ranking scores and the safety scores for the first candidate trajectory and the second candidate trajectory. 10. The method of claim 9, wherein the classification of the first and second potential trajectories is performed in parallel with the ranking of the first potential trajectory and the second potential trajectory. 11. The method of claim 9, wherein the classification of the first and second potential trajectories is performed independently of the ranking of the first potential trajectory and the second potential trajectory. 12. The method of claim 9, wherein the road section has two lanes and the first potential path contains a potential lane change maneuver, in response to determining by the electronic device that the first candidate path ranks higher than the second potential path and that the first potential path is safe, instructing includes the step of: require the electronic SDC device to attempt a potential lane change maneuver based on the first potential trajectory. 13. An electronic device for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the environment of the SDC, wherein the SDC is coupled to the electronic device, wherein the electronic device is configured to at a given time during SDC operation on a road section: generate a first candidate trajectory and a second candidate trajectory for the SDC on the road section, respectively; rank, using a machine learning ranking algorithm (MLA), the first candidate trajectory and the second candidate trajectory, wherein the ranking MLA is trained to predict a ranking score indicating the similarity between the corresponding candidate trajectory and the reference trajectory; classify, using the classifying MLA, the first potential trajectory and the second potential trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; identify the best trajectory among the first candidate trajectory and the second candidate trajectory and the safest trajectory among the first candidate trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; instruct the SDC to take the first action if the best trajectory and the safest trajectory are the same potential trajectory; And instruct the SDC to take a second action if the best trajectory and the safest trajectory are not the same potential trajectory. 14. The electronic device of claim 13, wherein the ranking MLA is a pairwise ranking algorithm and is trained on the first set of training data, wherein the first set of training data contains a reference trajectory representing a training trajectory manually executed by a human expert driving in the SDC on the training road section at the training time; And a first predicted training trajectory and a second predicted training trajectory generated by the electronic device for the SDC for the training road section at the training time; And the ranking MLA is trained on the first training data set to rank the first predicted training trajectory and the second predicted training trajectory based on the similarity between (i) the first predicted training trajectory and the second predicted training trajectory and (ii) the reference trajectory. 15. The electronic device of claim 13, wherein the classifying MLA is trained based on a second set of training data, wherein the second set of training data comprises: a training trajectory generated by the SDC electronic device on the training road section at the training time; a label indicating at least one of the implicit and explicit inputs of a human expert driving in the SDC on the training road section during the training time point. 16. The electronic device of claim 13, wherein the classifying MLA is a decision tree-based algorithm. 17. The electronic device of claim 13, wherein the road segment has two lanes and the first potential path includes a potential lane change maneuver, and the second potential path includes a potential lane keeping maneuver, and wherein: if both the best trajectory and the safest trajectory are the first potential trajectory, the first action is to attempt a potential lane change maneuver based on the first potential trajectory; if both the best trajectory and the safest trajectory are the second potential trajectory, the first action is to attempt a potential lane keeping maneuver based on the second potential trajectory. 18. The electronic device of claim 13, wherein the road section has two lanes of traffic and the first potential path comprises a potential lane change maneuver and the second potential path comprises a potential lane keeping maneuver and wherein, if the best trajectory is the first potential trajectory and the safest trajectory is the second potential trajectory, the second action is to turn on the visual SDC indicator to indicate the intent to perform the potential lane change maneuver while executing the potential lane keeping maneuver until another predetermined time after the predetermined time. 19. The electronic device of claim 13, wherein the road section has two lanes of traffic and the first potential path comprises a potential lane change maneuver and the second potential path comprises a potential lane keeping maneuver and wherein, if the best trajectory is the second potential trajectory and the safest trajectory is the first potential trajectory, the second action is to attempt a potential lane change maneuver based on the first potential trajectory. 20. Electronic device according to claim 18, wherein the electronic device is additionally configured to at another given point in time after a given point in time: identify a new best trajectory among the first potential trajectory and a second potential trajectory and a new safest trajectory among the first potential trajectory and a second potential trajectory based on a combination of the respective new ranking scores and the new safety scores; in response to the new best trajectory and the new safest trajectory, which are both the first potential trajectory, direct the SDC to perform a potential lane change maneuver based on the first potential trajectory; in response to the new best trajectory and the new safest trajectory, which are both the second potential trajectory, instruct SDC to perform a potential lane keeping maneuver based on the second potential trajectory. 21. A method for controlling a self-driving vehicle (SDC) on a stretch of road, wherein a sensor system is mounted on the SDC to collect information about the SDC's surroundings, wherein the SDC is coupled to an electronic device, the method including the steps of: at a given time during SDC operation on a road section: generating, using an electronic device, a first potential trajectory and a second potential trajectory for the SDC on the road section, respectively; ranking, using an electronic device using a machine learning ranking algorithm (MLA), the first potential trajectory and the second potential trajectory, wherein the ranking MLA is trained to predict a ranking score indicating the similarity between the corresponding candidate trajectory and the reference trajectory; classifying by an electronic device using a classifying MLA the first potential trajectory and the second potential trajectory, wherein the classifying MLA is trained to generate a safety score indicating whether the corresponding candidate trajectory is safe or not; electronically identifying the best trajectory among the first potential trajectory and the second potential trajectory and the safest trajectory among the first potential trajectory and the second potential trajectory based on a combination of the respective ranking scores and the safety scores; order, using the electronic SDC device, to work in accordance with the safest trajectory; if the safest path and the best path are not the same potential path, direct the SDC electronic device to perform a corrective action by operating in accordance with the safest path, the corrective action being intended to make the best path become the safest path in the future. 22. The method of claim 21, wherein the corrective action is to turn on the visual SDC indicator to indicate an intent to perform a potential lane change maneuver.