WO2022105437A1 - 一种路径规划方法、装置和电子设备 - Google Patents

一种路径规划方法、装置和电子设备 Download PDF

Info

Publication number
WO2022105437A1
WO2022105437A1 PCT/CN2021/121113 CN2021121113W WO2022105437A1 WO 2022105437 A1 WO2022105437 A1 WO 2022105437A1 CN 2021121113 W CN2021121113 W CN 2021121113W WO 2022105437 A1 WO2022105437 A1 WO 2022105437A1
Authority
WO
WIPO (PCT)
Prior art keywords
road
network
value
environmental
path
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/121113
Other languages
English (en)
French (fr)
Inventor
邹李兵
张一凡
宁越
张富强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Goertek Inc
Original Assignee
Goertek Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Goertek Inc filed Critical Goertek Inc
Priority to US18/037,914 priority Critical patent/US12492910B2/en
Publication of WO2022105437A1 publication Critical patent/WO2022105437A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/34—Route searching; Route guidance
    • G01C21/3453—Special cost functions, i.e. other than distance or default speed limit of road segments
    • G01C21/3492—Special cost functions, i.e. other than distance or default speed limit of road segments employing speed data or traffic data, e.g. real-time or historical
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/34—Route searching; Route guidance
    • G01C21/3407—Route searching; Route guidance specially adapted for specific applications
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G01C21/34—Route searching; Route guidance
    • G01C21/3446—Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags or using precalculated routes
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/092—Reinforcement learning
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W2554/00—Input parameters relating to objects
    • B60W2554/40—Dynamic objects, e.g. animals, windblown objects
    • B60W2554/404—Characteristics
    • B60W2554/4042—Longitudinal speed
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W2554/00—Input parameters relating to objects
    • B60W2554/40—Dynamic objects, e.g. animals, windblown objects
    • B60W2554/406—Traffic density

Definitions

  • the present application relates to the field of computer technology, and in particular, to a path planning method, apparatus and electronic device.
  • Path planning is the basis for realizing navigation and even automatic driving.
  • the commonly used methods of path planning mainly include static path planning and dynamic path planning.
  • Static path planning is suitable for environments where agents (such as agents) and tasks are relatively unchanged. This scenario is an ideal scenario, but in real life, it is more of a dynamic scenario, and its dynamic elements include individual agents.
  • the number is dynamic, the speed of the agent is variable, the randomness of the task, and the influence of various environmental noises, so dynamic path planning has greater practicability and greater challenges.
  • dynamic path planning is mainly based on single-body dynamic planning (such as D*), but because the influence of other autonomously movable agents is not considered, the effect is not good in a multi-body dynamic environment.
  • path planning methods such as DWA (Dynamic Window Approach, dynamic window method) have a small time window and have time limitations.
  • the embodiments of the present application provide a path planning method, apparatus, and electronic device, so as to realize the spatial globality and the temporal globality of the path planning.
  • an embodiment of the present application provides a path planning method, which includes: performing environmental modeling according to static road network information and dynamic road condition information of a road network to obtain an environment model, where the dynamic road condition information includes data of multiple agents in the road network. motion state; determine multiple candidate paths according to the starting point and end point; use the feature extraction network of the path planning model to extract the environmental features corresponding to each candidate path from the environmental model; input the environmental features into the value estimation network of the path planning model to obtain The value estimates the estimated value of each candidate path output by the network; determines the optimal path among the candidate paths according to the estimated value.
  • an embodiment of the present application further provides a path planning device, the device comprising:
  • an environment modeling unit configured to perform environment modeling according to static road network information and dynamic road condition information of the road network to obtain an environment model, where the dynamic road condition information includes motion states of multiple agents in the road network;
  • the candidate path unit is used to determine multiple candidate paths according to the starting point and the ending point;
  • the feature extraction unit is used for extracting the environment features corresponding to each candidate path from the environment model by using the feature extraction network of the path planning model;
  • the value estimation unit is used to input the environmental characteristics into the value estimation network of the path planning model, and obtain the estimated value of each candidate path output by the value estimation network;
  • the path determination unit is used to determine the optimal path among the candidate paths according to the estimated value.
  • embodiments of the present application further provide an electronic device, including: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to execute the above path planning method.
  • embodiments of the present application further provide a computer-readable storage medium, where the computer-readable storage medium stores one or more programs, and when the one or more programs are executed by an electronic device including multiple application programs, The device performs the above path planning method.
  • both static and dynamic information are considered during environment modeling, so that the practicability of path planning is higher, and the dynamic road condition information includes information in the road network.
  • the motion states of multiple agents are spatially global; during path planning, according to the environment model, the environmental features corresponding to the candidate paths determined according to the starting point and the end point are extracted, thus realizing the planning of the whole path instead of dividing the time.
  • the planning of the window can balance the benefits of the current decision and the total benefits in the future, and it is global in time.
  • FIG. 1 shows a schematic flowchart of a path planning method according to an embodiment of the present application
  • Fig. 2 shows a schematic flowchart of environmental feature extraction of candidate paths according to an embodiment of the present application
  • FIG. 3 shows a schematic flowchart of a path planning method according to an embodiment of the present application
  • FIG. 4 is a schematic structural diagram of an electronic device in an embodiment of the present application.
  • the technical idea of the present application is to consider not only the state of itself, but also the states of other agents when performing path planning, and to plan the path as a whole, thereby taking into account the globality of both space and time.
  • FIG. 1 shows a schematic flowchart of a path planning method according to an embodiment of the present application. As shown in Figure 1, the method includes:
  • Step S110 Perform environmental modeling according to static road network information and dynamic road condition information of the road network to obtain an environmental model, and the dynamic road condition information includes motion states of multiple agents in the road network.
  • Road network is the abbreviation of road traffic network, and static road network information can describe static and invariable information such as road direction, connection relationship, and length.
  • static road network information can describe static and invariable information such as road direction, connection relationship, and length.
  • other agents are usually driving on the road.
  • the present application also uses dynamic road condition information in environment modeling to consider the motion states of multiple agents in the road network.
  • an agent refers to a target that can move autonomously in a road network, such as various types of vehicles.
  • Step S120 Determine a plurality of candidate paths according to the starting point and the ending point.
  • the process of determining the candidate path can be implemented by using the prior art, for example, selecting any path finding algorithm. According to a pair of starting point and ending point, multiple candidate paths can usually be found, and the goal of this application is to select the optimal path from the multiple candidate paths.
  • step S120 and step S110 are not strictly executed sequentially, and may also be executed in parallel.
  • Step S130 using the feature extraction network of the path planning model to extract the environment features corresponding to each candidate path from the environment model.
  • step S140 the environmental characteristics are input into the value estimation network of the path planning model, and the estimated value of each candidate path output by the value estimation network is obtained.
  • the path planning model includes two sub-networks, namely, a feature extraction network for environmental post extraction and a value estimation network for path comparison.
  • the estimated value can be realized by selecting different dimensions according to actual needs. For example, if the time dimension is selected, the estimated value is the predicted travel time; in addition, the safety dimension, comfort dimension, etc. can also be selected, which will not be listed here.
  • Step S150 Determine the optimal path among the candidate paths according to the estimated value.
  • This step can be implemented using existing technology, such as the E-greedy algorithm.
  • the method shown in Fig. 1 considers both static and dynamic information when modeling the environment, which makes the path planning more practical, and the dynamic road condition information includes the motion states of multiple agents in the road network. It has spatial globality; during path planning, according to the environment model, the environmental characteristics corresponding to the candidate paths determined according to the starting point and the end point are extracted, so as to realize the planning of the whole path rather than the planning of time windows, which can balance the current decision-making
  • the revenue and future total revenue are global in time.
  • using the feature extraction network of the path planning model to extract the environmental features corresponding to each candidate path from the environment model includes: extracting the static environmental features and dynamic characteristics of the road segments from the environment model for the road segments included in the candidate routes Environmental characteristics, the static environmental characteristics and dynamic environmental characteristics of the road section are spliced to obtain the environmental characteristics of the road section; the long short-term memory neural network LSTM is used to determine the environmental characteristics of the candidate path according to the environmental characteristics of each road section in the candidate path.
  • a candidate path from the starting point to the ending point is usually composed of multiple road segments, each of which is a road segment.
  • the environmental characteristics of each road segment can be determined first. Since the environment modeling includes both dynamic and static aspects, the static and dynamic environmental features of the road section can be extracted separately, and then the two can be spliced to obtain the environmental characteristics of the road section.
  • the concat() function can be used to implement the splicing operation.
  • the present application uses LSTM to obtain the environmental characteristics of the candidate paths based on the obtained environmental characteristics of the road sections, so as to ensure the uniform format of the environmental characteristics of the candidate paths, and realize the candidate path embedding (Path-Embedding).
  • the static road network information includes a vector representation of road segments
  • the static environmental features and dynamic environmental features of the road section include: using the graph neural network Graph-Bert to extract feature vectors from the directed graph as the static environmental features of the road section.
  • the route planning uses a raster road network map obtained by rasterizing the route.
  • the grid road network map reduces the accuracy because the continuous road network is discretized into each road network grid, which is a compromise for the convenience of neural network learning.
  • the vector representation of the road segment is directly used to establish a directed graph of the road network instead of the grid map of the road network.
  • V is a set of vertices
  • V is a set of road segments
  • each road segment in the road network is represented by its elements.
  • E is a set of relationships between road segments.
  • W is the set of road segment lengths, represents the length of a road segment li .
  • the connection relationship between the nodes of the directed graph is extracted through the Graph-Bert network, and the feature vector of the specified road segment l in the road segment set V is output through the query operation.
  • the Graph-Bert network is a graph neural network with Bert (Bidirectional Encoder Representations from Transformers, Transformer-based bidirectional encoder representation) added, which can realize the transfer of directed graphs through the Attention mechanism.
  • Bert Bidirectional Encoder Representations from Transformers, Transformer-based bidirectional encoder representation
  • the correlation of the nodes of the directed graph G can be extracted for node feature expression or directed graph reconstruction.
  • performing environmental modeling according to static road network information and dynamic road condition information of the road network, and obtaining the environmental model includes: determining dynamic environmental attributes of each road section according to the dynamic road condition information, and the dynamic environmental attributes include the dimension of the number of agents and/or Or the speed dimension; using the feature extraction network of the path planning model to extract the environmental features corresponding to each candidate path from the environmental model includes: using the values of each dimension of the dynamic environmental attributes to splicing into a feature vector as a dynamic environmental feature, or using neural The network extracts feature vectors from dynamic environment attributes as dynamic environment features.
  • the dynamic environment feature can be used as the dynamic environment feature by splicing the values of each dimension of the dynamic environment attribute into a feature vector, easy to use.
  • a neural network can also be used to extract feature vectors from dynamic environment attributes as dynamic environment features, but the time required for path planning will increase accordingly.
  • the average speed of the agent can be calculated as the value of the speed dimension, which means that when planning the path, it can complete the segment of the candidate path at the average speed of the agent.
  • the average speed of the agent (this time is 0) is still used as the value of the speed dimension, it will cause itself to be unable to pass in this road section.
  • the road section is an ideal situation and can pass at the maximum speed limit of the road section. That is, in some embodiments, if the value of the number dimension of agents is not 0, the value of the speed dimension is the average speed of the agent; if the value of the number dimension of agents is 0, the value of the speed dimension is the preset maximum speed .
  • the value of the speed dimension represents the speed that the road segment can adopt during actual driving.
  • the estimated value is the predicted transit time
  • the method further includes: using each candidate path as a training sample, and using the predicted transit time of the candidate path as the sample predicted value, obtaining the simulated transit time and/or actual transit time of the candidate path
  • the transit time is taken as the real value of the sample
  • the training loss value is determined according to the predicted value of the sample and the real value of the sample
  • the parameters of the feature extraction network and the time prediction network are updated according to the training loss value and the back-propagation algorithm.
  • a simulator can be used to simulate the vehicle traveling on the planned path to obtain the simulated transit time.
  • determining the training loss value according to the sample predicted value and the sample real value includes: calculating the training loss value using a mean square error function where y is the true value of the sample, is the sample predicted value, and m is the number of samples.
  • the mean square error function can calculate the Euclidean distance between the predicted value of the sample and the true value of the sample. The closer the sample predicted value and the sample real value are, the smaller the mean square error of the two, and the better the effect of the path planning model.
  • using each candidate path as a training sample includes: putting the training sample into the sample experience pool; when the number of training samples in the sample experience pool reaches a preset value, executing the training loss value and reverse
  • the propagation algorithm updates the parameters of the feature extraction network and the time prediction network, and initializes the sample experience pool after the parameters are updated. In this way, online learning of the path planning model can be realized, and the path planning model can be updated in time to improve the effect.
  • Step S210 initialize the experience pool D, and set its capacity to N.
  • Step S220 initialize the feature extraction network F-Network of the path planning model, and assign random values to its parameters.
  • step S230 the value estimation network Q-Network of the path planning model is initialized, and its parameters are randomly assigned.
  • Step S240 according to the static road network information, generate a directed graph G, and initialize Graph-Bert.
  • Step S250 using multiple training stages to train the path planning model. Specifically, the following sub-steps S251 to S253 are performed in each training stage:
  • Step S251 simulate and generate a plurality of path planning tasks T(start, end) including the start point start and the end point end, and the number of path planning tasks is M.
  • Array B is initialized with a value of 0.
  • Step S253, M agents are generated in parallel. Each agent further performs the following sub-steps S2531-S2536:
  • Step S2531 for each task T, find candidate paths ⁇ p 0 , p 1 , p 2 , . . . , p n ⁇ through the path search algorithm.
  • Step S2532 for the candidate path p i , referring to the method shown in FIG. 2 , use F-Network to extract the environmental features of the candidate path, which specifically includes sub-steps S25321 to S25324:
  • Step S25321 using the graph neural network Graph-Bert to extract the feature vector z s from the directed graph as the static environment feature of the road segment.
  • Step S25322 determine the road section
  • the dynamic environment feature za of is ( avgSpeed , carNums), and the specific determination method may refer to the foregoing embodiment.
  • Step S25323 splicing z s and z a into road sections through the concat() function environmental characteristics.
  • FC Flexible Connected, fully connected
  • Step S25324 will Input into LSTM to get the environmental features of the candidate path pi
  • Step S2533 the Input to Q-Network to get predicted transit time
  • Step S2534 using E-greedy algorithm according to The optimal path p k is selected.
  • Step S2535 simulate running the agent, and update the dynamic array B of the road network.
  • Step S2536 record the simulated transit time y k to obtain a training sample Added to sample experience pool D.
  • Step S254 if the number of new samples in the sample experience pool is greater than 0.7*N, calculate the mean square error loss of these new samples, and use the back propagation algorithm to carry out the parameters of F-Network and Q-Network according to the calculated mean square error loss. renew.
  • Step S255 reset the road network dynamic array B.
  • step S253 In the process of online learning, it is only necessary to replace the agent in step S253 and its sub-steps with an actual vehicle, that is, the actual running time can be obtained according to the actual operation of the vehicle without simulating the running of the agent.
  • the embodiments of the present application also provide a path planning apparatus, which can be used to implement the path planning method shown in any of the above embodiments.
  • FIG. 3 shows a schematic structural diagram of a path planning apparatus according to an embodiment of the present application.
  • the path planning apparatus 300 includes:
  • the environment modeling unit 310 is configured to perform environment modeling according to static road network information and dynamic road condition information of the road network to obtain an environment model, and the dynamic road condition information includes motion states of multiple agents in the road network.
  • the candidate path unit 320 is configured to determine a plurality of candidate paths according to the start point and the end point.
  • the feature extraction unit 330 is configured to extract environmental features corresponding to each candidate path from the environment model by using the feature extraction network of the path planning model.
  • the value estimation unit 340 is configured to input the environmental characteristics into the value estimation network of the path planning model, and obtain the estimated value of each candidate path output by the value estimation network.
  • the path determination unit 350 is configured to determine the optimal path among the candidate paths according to the estimated value.
  • the feature extraction unit 330 is configured to extract the static environment feature and dynamic environment feature of the road segment from the environment model for the road segment included in the candidate route, and splicing the static environment feature and dynamic environment feature of the road segment to obtain The environmental characteristics of the road segment; the long short-term memory neural network LSTM is used to determine the environmental characteristics of the candidate route according to the road segment environmental characteristics of each road segment in the candidate route.
  • the static road network information includes a vector representation of road segments
  • the environment modeling unit 310 is configured to determine the dynamic environment attributes of each road segment according to the dynamic road condition information, and the dynamic environment attributes include the dimension of the number of agents and/or the speed dimension; the feature extraction unit 330 is configured to utilize the dynamic environment The value of each dimension of the attribute is spliced into a feature vector as a dynamic environment feature, or a neural network is used to extract a feature vector from the dynamic environment attribute as a dynamic environment feature.
  • the value of the speed dimension is the average speed of the agent; if the value of the number of agents is 0, the value of the speed dimension is the preset maximum speed.
  • the estimated value is the predicted transit time
  • the apparatus further includes a training unit configured to use each candidate path as a training sample, and use the predicted transit time of the candidate path as the sample predicted value to obtain the simulated transit time of the candidate path and/or the actual transit time as the sample real value; the training loss value is determined according to the sample predicted value and the sample real value; the parameters of the feature extraction network and the time prediction network are updated according to the training loss value and the back propagation algorithm.
  • a training unit for computing a training loss value using a mean square error function where y is the true value of the sample, is the sample predicted value, and m is the number of samples.
  • the training unit is configured to put the training samples into the sample experience pool; in the case that the number of training samples in the sample experience pool reaches a preset value, perform the feature analysis according to the training loss value and the back-propagation algorithm.
  • the steps of extracting the parameters of the network and the temporal prediction network for updating, and initializing the sample experience pool after the parameters are updated.
  • FIG. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
  • the electronic device includes a processor, and optionally an internal bus, a network interface, and a memory.
  • the memory may include memory, such as high-speed random-access memory (Random-Access Memory, RAM), and may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
  • RAM random-Access Memory
  • non-volatile memory such as at least one disk memory.
  • the electronic equipment may also include hardware required for other services.
  • the processor, network interface and memory can be connected to each other through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect, peripheral component interconnect standard) bus or an EISA (Extended Component Interconnect) bus. Industry Standard Architecture, extended industry standard structure) bus, etc.
  • the bus can be divided into address bus, data bus, control bus and so on. For ease of presentation, only one bidirectional arrow is used in FIG. 4, but it does not mean that there is only one bus or one type of bus.
  • the program may include program code, and the program code includes computer operation instructions.
  • the memory may include memory and non-volatile memory and provide instructions and data to the processor.
  • the processor reads the corresponding computer program from the non-volatile memory into the memory and runs it, forming a path planning device on the logical level.
  • the processor executes the program stored in the memory, and is specifically used to perform the following operations:
  • the dynamic road condition information includes the motion states of multiple agents in the road network; determine multiple candidate paths according to the starting point and end point; use the path planning model It extracts the environmental features corresponding to each candidate path from the environmental model; inputs the environmental features into the value estimation network of the path planning model, and obtains the estimated value of each candidate path output by the value estimation network; according to the estimated value Determine the optimal path among the candidate paths.
  • the above-mentioned method executed by the path planning apparatus disclosed in the embodiment shown in FIG. 1 of the present application may be applied to a processor, or implemented by a processor.
  • a processor may be an integrated circuit chip with signal processing capabilities.
  • each step of the above-mentioned method can be completed by a hardware integrated logic circuit in a processor or an instruction in the form of software.
  • the above-mentioned processor can be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), etc.; it can also be a digital signal processor (Digital Signal Processor, DSP), dedicated integrated Circuit (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the steps of the methods disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor.
  • the software module may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art.
  • the storage medium is located in the memory, and the processor reads the information in the memory, and completes the steps of the above method in combination with its hardware.
  • the electronic device can also execute the method executed by the path planning apparatus in FIG. 1 , and implement the functions of the path planning apparatus in the embodiment shown in FIG. 1 .
  • the electronic device can also execute the method executed by the path planning apparatus in FIG. 1 , and implement the functions of the path planning apparatus in the embodiment shown in FIG. 1 .
  • For other specific functions performed by the electronic device in this embodiment reference may be made to the relevant content of the foregoing method embodiments, and details are not described herein again.
  • the embodiments of the present application also provide a computer-readable storage medium, where the computer-readable storage medium stores one or more programs, and the one or more programs include instructions, and the instructions are executed by an electronic device including multiple application programs.
  • the electronic device can be made to execute the method executed by the path planning apparatus in the embodiment shown in FIG. 1 , and is specifically used to execute:
  • the dynamic road condition information includes the motion states of multiple agents in the road network; determine multiple candidate paths according to the starting point and end point; use the path planning model It extracts the environmental features corresponding to each candidate path from the environmental model; inputs the environmental features into the value estimation network of the path planning model, and obtains the estimated value of each candidate path output by the value estimation network; according to the estimated value Determine the optimal path among the candidate paths.
  • the embodiments of the present application may be provided as a method, a system, or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
  • computer-usable storage media including, but not limited to, disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions
  • the apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • Computer-readable media includes both persistent and non-permanent, removable and non-removable media, and storage of information may be implemented by any method or technology.
  • Information may be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Flash Memory or other memory technology, Compact Disc Read Only Memory (CD-ROM), Digital Versatile Disc (DVD) or other optical storage, Magnetic tape cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.
  • computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Automation & Control Theory (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Traffic Control Systems (AREA)

Abstract

一种路径规划方法、装置和电子设备。其中方法包括：根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型(S110)；根据起点和终点确定多条候选路径(S120)；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征(S130)；将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值(S140)；根据预估价值确定各候选路径中的最优路径(S150)。该动态路况信息中包括路网中多个智能体的运动状态，具有空间上的全局性；在进行路径规划时，根据环境模型，提取根据起点和终点确定的候选路径对应的环境特征，从而实现了全路径的规划而非分时间窗口的规划，具有时间上的全局性。

Description

一种路径规划方法、装置和电子设备 技术领域
本申请涉及计算机技术领域，尤其涉及一种路径规划方法、装置和电子设备。
发明背景
路径规划是实现导航乃至自动驾驶的基础，目前路径规划常用的方法主要包括静态路径规划和动态路径规划这两大类。静态路径规划适合智能体(例如智能体)和任务相对不变的环境下，这种场景是理想状态下的场景，但在实际的生活中，更多是动态场景，其动态要素包括智能体个数是动态的、智能体速度是可变的、任务的随机性以及各种环境噪声的影响，因而动态路径规划具有更大的实用性，也具有更大的挑战性。
现有技术中，动态路径规划主要是以单体动态规划为主(如D*)，但由于未考虑其他可自主移动的智能体的影响，在多体动态环境下，效果并不好。另外如DWA(Dynamic Window Approach，动态窗口法)等路径规划方法，时间窗口小，具有时间上的局限性。
由此，需要一种改进的路径规划方式。
发明内容
本申请实施例提供了一种路径规划方法、装置和电子设备，以实现路径规划的空间全局性和时间全局性。
本申请实施例采用下述技术方案：
第一方面，本申请实施例提供一种路径规划方法，包括：根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态；根据起点和终点确定多条候选路径；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；根据预估价值确定各候选路径中的最优路径。
第二方面，本申请实施例还提供一种路径规划装置，该装置包括：
环境建模单元，用于根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，所述动态路况信息包括路网中多个智能体的运动状态；
候选路径单元，用于根据起点和终点确定多条候选路径；
特征提取单元，用于利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；
价值估算单元，用于将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；
路径确定单元，用于根据预估价值确定各候选路径中的最优路径。
第三方面，本申请实施例还提供一种电子设备，包括：处理器；以及被安排成存储计算机可执行指令的存储器，可执行指令在被执行时使处理器执行如上的路径规划方法。
第四方面，本申请实施例还提供一种计算机可读存储介质，计算机可读存储介质存储一个或多个程序，一个或多个程序当被包括多个应用程序的电子设备执行时，使得电子设备执行如上的路径规划方法。
本申请实施例采用的上述至少一个技术方案能够达到以下有益效果：在进行环境建模时考虑了静态和动态两方面信息，使路径规划的实用性更高，并且动态路况信息中包括路网中多个智能体的运动状态，具有空间上的全局性；在进行路径规划时，根据环境模型，提取根据起点和终点确定的候选路径对应的环境特征，从而实现了全路径的规划而非分时间窗口的规划，能够平衡当前决策的收益和未来的总收益，具有时间上的全局性。
附图简要说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1示出了根据本申请一个实施例的路径规划方法的流程示意图；
图2示出了根据本申请一个实施例的候选路径的环境特征提取的流程示意 图；
图3示出了根据本申请一个实施例的路径规划方法的流程示意图；
图4为本申请实施例中一种电子设备的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚，下面将结合本申请具体实施例及相应的附图对本申请技术方案进行清楚、完整地描述。显然，所描述的实施例仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
本申请的技术构思在于，在进行路径规划时不仅考虑自身状态，还考虑了其他智能体的状态，并且对路径进行整体规划，由此兼顾了空间和时间双方面的全局性。
以下结合附图，详细说明本申请各实施例提供的技术方案。
图1示出了根据本申请一个实施例的路径规划方法的流程示意图。如图1所示，该方法包括：
步骤S110，根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态。
路网是道路交通网络的简称，静态路网信息能够描述道路方向、连接关系、长度等静态不变的信息。而实际场景中，道路上通常行驶着其他智能体，为了提升技术方案的实用价值，本申请还在环境建模时使用了动态路况信息，以考虑路网中多个智能体的运动状态。
本申请中，智能体所指代的是在可以在路网中自主移动的目标，例如各类车辆。
步骤S120，根据起点和终点确定多条候选路径。
确定候选路径的过程可以利用现有技术，例如选用任一种路径查找算法来实现。根据一对起点和终点，通常可以查找到多条候选路径，本申请的目标就 是从多条候选路径中选出最优路径。
需要说明的是，步骤S120和步骤S110没有严格的先后执行顺序，也可以并行执行。
步骤S130，利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征。
步骤S140，将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值。
本申请中，路径规划模型包括两个子网络，即用于进行环境帖子提取的特征提取网络和进行路径比较的价值估算网络。预估价值可以根据实际需求选择不同维度来实现，例如选择时间维度，则预估价值就是预测通行时间；此外还可以选择安全维度、舒适性维度等等，在此不再一一列举。
步骤S150，根据预估价值确定各候选路径中的最优路径。
该步骤可以利用现有技术，如E-greedy算法来实现。
可见，图1所示的方法，在进行环境建模时考虑了静态和动态两方面信息，使路径规划的实用性更高，并且动态路况信息中包括路网中多个智能体的运动状态，具有空间上的全局性；在进行路径规划时，根据环境模型，提取根据起点和终点确定的候选路径对应的环境特征，从而实现了全路径的规划而非分时间窗口的规划，能够平衡当前决策的收益和未来的总收益，具有时间上的全局性。
在一些实施例中，利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征包括：对候选路径中包含的路段，从环境模型中提取路段的静态环境特征和动态环境特征，将路段的静态环境特征和动态环境特征进行拼接，得到路段环境特征；利用长短期记忆神经网络LSTM，根据候选路径中各路段的路段环境特征，确定候选路径的环境特征。
从起点到终点的候选路径，通常是由多个路段组成的，每个路段是一段道路。为了得到候选路径的环境特征，可以先确定每个路段的环境特征。由于环境建模包括动态和静态两方面，因此可以分别提取路段的静态环境特征和动态 环境特征，再将二者拼接得到路段的环境特征，拼接操作可以具体利用concat()函数来实现。
再得到各路段的环境特征后，会面对一个问题：对于多个候选路径来说，各候选路径所包含的路段数量和组成不完全一致(否则就构成了相同路径)，如果进行简单的拼接操作，就会造成各候选路径的环境特征长度不一致，这样不便于后续进行价值估算。因此，本申请基于已得到的路段的环境特征，利用LSTM得到候选路径的环境特征，从而确保候选路径的环境特征的格式统一，实现了候选路径嵌入(Path-Embedding)。
在一些实施例中，静态路网信息包括路段的矢量表示，根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型包括：将静态路网信息构建成路网有向图G＝(V，E，W)，其中V中的元素表征路网中的各路段，E中的元素表征各路段之间的关系，W中的元素表征各路段的长度；从环境模型中提取路段的静态环境特征和动态环境特征包括：利用图神经网络Graph-Bert，从有向图中提取特征向量作为路段的静态环境特征。
在一些技术方案中，路径规划使用的是对路径经过栅格化处理得到的栅格路网地图。栅格路网地图由于是把连续的路网离散化为一个个路网栅格，因此降低了精度，是一种为了便于神经网络学习的折中方案。在本申请的实施例中，则是直接利用路段的矢量表示，建立路网有向图来代替路网栅格地图。
在有向图中，V是顶点集，而在本申请的实施例中将V作为路段集合，以其中的元素表征路网中的各路段。E是路段与路段之间的关系集合，在一些实施例中，对于e ij{e ij∈E}取值如下(e ij表征路段l i与路段l j之间的关系)：若路段l i与路段l j不邻接，则e ij＝0；若路段l i与路段l j邻接，且从路段l i入路段l j，则e ij＝1；若路段l i与路段l j邻接，且从路段从l j入路段l i，则e ij＝2；e ii＝3。需要说明的是，e ij的具体取值可以不限于上面给出的示例。
W是路段长度集合，
Figure PCTCN2021121113-appb-000001
表示某路段l i的长度。
构建有向图G后，通过Graph-Bert网络提取有向图的节点之间的连接关系， 通过查询操作输出路段集合V中指定路段l的特征向量。
Graph-Bert网络是加入了Bert(Bidirectional Encoder Representations from Transformers，基于Transformer的双向编码器表示)的图神经网络，可以通过注意力Attention机制，实现有向图的迁移(transfer)。通过Graph-Bert可提取有向图G的节点的相关性，用于节点特征表达或有向图的重构。
在一些实施例中，根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型包括：根据动态路况信息确定各路段的动态环境属性，动态环境属性包括智能体数量维度和/或速度维度；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征包括：利用动态环境属性的各维度的值拼接成特征向量作为动态环境特征，或者，利用神经网络从动态环境属性中提取特征向量作为动态环境特征。
在一些实施例中，考虑到动态路况信息对路径规划的影响因素主要在于路网中智能体的数量和速度，因此可以通过将动态环境属性的各维度的值拼接成特征向量作为动态环境特征，操作简单。当然，如果为了提升精度，也可以利用神经网络从动态环境属性中提取特征向量作为动态环境特征，但是相应地会增加路径规划所需的时间。
具体地，可以计算智能体平均速度作为速度维度的值，这意味着在进行路径规划时，自身可以以智能体平均速度完成候选路径的该路段。然而在特殊情况下也会出现问题，那就是某一路段没有智能体，那么如果仍然是将智能体平均速度(此时为0)作为速度维度的值，就会导致自身在该路段无法通行。而实际上，若该路段没有其他智能体，意味着该路段是较为理想的情况，可以以该路段的最大限速通行。即在一些实施例中，若智能体数量维度的值不为0，则速度维度的值为智能体平均速度；若智能体数量维度的值为0，则速度维度的值为预设的最大速度。
例如，路段l的动态环境特征表示为向量(avgSpeed，carNums)，其中，carNums是路段l上的车辆总数，若carNums不为0，则
Figure PCTCN2021121113-appb-000002
CarSpeed i是第i辆车的速度，n＝carNums，若carNums＝0，则avgSpeed＝maxSpeed，maxSpeed为预设的最大速度，表征路段l的最高限速。
可以看出，速度维度的值表征路段在实际行驶时可采用的速度。
在一些实施例中，预估价值为预测通行时间，该方法还包括：将各候选路径作为训练样本，将候选路径的预测通行时间作为样本预测值，获取候选路径的模拟通行时间和/或实际通行时间作为样本真实值；根据样本预测值和样本真实值确定训练损失值；根据训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新。
在离线场景下由于无法获取实际通行时间，可以利用模拟器模拟车辆在规划的路径上行驶，得到模拟通行时间。
在一些实施例中，根据样本预测值和样本真实值确定训练损失值包括：利用均方差函数计算训练损失值
Figure PCTCN2021121113-appb-000003
其中y为样本真实值，
Figure PCTCN2021121113-appb-000004
为样本预测值，m为样本数量。均方差函数可以计算出样本预测值和样本真实值之间的欧式距离。样本预测值和样本真实值越接近，两者的均方差就越小，路径规划模型的效果也就越好。
在一些实施例中，将各候选路径作为训练样本包括：将训练样本放入样本经验池中；在样本经验池中的训练样本数量达到预设值的情况下，执行根据训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新的步骤，并在参数更新后，对样本经验池进行初始化。由此可以实现路径规划模型的在线学习，能够及时更新路径规划模型以提升效果。
下面以一个具体实施例进行介绍。
步骤S210，初始化经验池D，设置其容量大小为N。
步骤S220，初始化路径规划模型的特征提取网络F-Network，对其参数进行随机赋值。
步骤S230，初始化路径规划模型的价值估算网络Q-Network，对其参数 进行随机赋值。
步骤S240，根据静态路网信息，生成有向图G，初始化Graph-Bert。
步骤S250，利用多个训练阶段对路径规划模型进行训练。具体地，在每个训练阶段中执行如下的子步骤S251～S253：
步骤S251，模拟生成多个包含起点start和终点end的路径规划任务T(start，end)，路径规划任务个数为M。
步骤S252，初始化路网动态数组B，数组长度与有向图G中V的大小相等，即Size(D)＝Size(V)。数组B初始化值为0。
步骤S253，并行生成M个agent。每个agent进一步执行如下的子步骤S2531～S2536：
步骤S2531，对每个任务T，通过路径查找算法，找出候选路径{p 0，p 1，p 2，...，p n}。
步骤S2532，对候选路径p i，参照图2示出的方式，利用F-Network进行候选路径的环境特征的提取，具体包括子步骤S25321～S25324：
步骤S25321，利用图神经网络Graph-Bert，从有向图中提取特征向量z s作为路段的静态环境特征。
步骤S25322，确定路段
Figure PCTCN2021121113-appb-000005
的动态环境特征z a为(avgSpeed，carNums)，具体确定方式可以参照前述实施例。
步骤S25323，通过concat()函数将z s和z a拼接成路段
Figure PCTCN2021121113-appb-000006
的环境特征。具体地，可以再通过两个FC(Fully Connected，全连接)层对拼接的特征进行处理，得到路段
Figure PCTCN2021121113-appb-000007
的环境特征
Figure PCTCN2021121113-appb-000008
步骤S25324，将
Figure PCTCN2021121113-appb-000009
输入到LSTM中，得到候选路径p i的环境特征
Figure PCTCN2021121113-appb-000010
步骤S2533，将
Figure PCTCN2021121113-appb-000011
输入到Q-Network，得到预测通行时间
Figure PCTCN2021121113-appb-000012
步骤S2534，利用E-greedy算法根据
Figure PCTCN2021121113-appb-000013
选择出最优路径p k。
步骤S2535，模拟运行agent，并更新路网的动态数组B。
步骤S2536，记录模拟通行时间y k，得到一个训练样本
Figure PCTCN2021121113-appb-000014
添加到 样本经验池D。
步骤S254，若样本经验池中的新样本数>0.7*N，则计算这些新样本的均方差损失，根据计算的均方差损失，利用反向传播算法对F-Network和Q-Network的参数进行更新。
步骤S255，重置路网动态数组B。
在进行在线学习的过程中，只需要将步骤S253及其子步骤中的agent用实际车辆代替即可，即不需要模拟运行agent而是可以根据车辆的实际运行情况得到实际通行时间。
本申请的实施例还提供了一种路径规划装置，可用于实现如上任一实施例所示的路径规划方法。
具体地，图3示出了根据本申请一个实施例的路径规划装置的结构示意图。如图3所示，路径规划装置300包括：
环境建模单元310，用于根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态。
候选路径单元320，用于根据起点和终点确定多条候选路径。
特征提取单元330，用于利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征。
价值估算单元340，用于将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值。
路径确定单元350，用于根据预估价值确定各候选路径中的最优路径。
在一些实施例中，特征提取单元330，用于对候选路径中包含的路段，从环境模型中提取路段的静态环境特征和动态环境特征，将路段的静态环境特征和动态环境特征进行拼接，得到路段环境特征；利用长短期记忆神经网络LSTM，根据候选路径中各路段的路段环境特征，确定候选路径的环境特征。
在一些实施例中，静态路网信息包括路段的矢量表示，环境建模单元310，用于将静态路网信息构建成路网有向图G＝(V，E，W)，其中V中的元素表征路网中的各路段，E中的元素表征各路段之间的关系，W中的元素表征各路段的长度； 特征提取单元330，用于利用图神经网络Graph-Bert，从有向图中提取特征向量作为路段的静态环境特征。
在一些实施例中，环境建模单元310，用于根据动态路况信息确定各路段的动态环境属性，动态环境属性包括智能体数量维度和/或速度维度；特征提取单元330，用于利用动态环境属性的各维度的值拼接成特征向量作为动态环境特征，或者，利用神经网络从动态环境属性中提取特征向量作为动态环境特征。
在一些实施例中，若智能体数量维度的值不为0，则速度维度的值为智能体平均速度；若智能体数量维度的值为0，则速度维度的值为预设的最大速度。
在一些实施例中，预估价值为预测通行时间，该装置还包括训练单元，用于将各候选路径作为训练样本，将候选路径的预测通行时间作为样本预测值，获取候选路径的模拟通行时间和/或实际通行时间作为样本真实值；根据样本预测值和样本真实值确定训练损失值；根据训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新。
在一些实施例中，训练单元，用于利用均方差函数计算训练损失值
Figure PCTCN2021121113-appb-000015
其中y为样本真实值，
Figure PCTCN2021121113-appb-000016
为样本预测值，m为样本数量。
在一些实施例中，训练单元，用于将训练样本放入样本经验池中；在样本经验池中的训练样本数量达到预设值的情况下，执行根据训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新的步骤，并在参数更新后，对样本经验池进行初始化。
能够理解，上述路径规划装置，能够实现前述实施例中提供的路径规划方法的各个步骤，关于路径规划方法的相关阐释均适用于路径规划装置，此处不再赘述。
图4是本申请的一个实施例电子设备的结构示意图。请参考图4，在硬件层面，该电子设备包括处理器，可选地还包括内部总线、网络接口、存储器。其中，存储器可能包含内存，例如高速随机存取存储器(Random-Access Memory，RAM)，也可能还包括非易失性存储器(non-volatile memory)，例如至少1个磁 盘存储器等。当然，该电子设备还可能包括其他业务所需要的硬件。
处理器、网络接口和存储器可以通过内部总线相互连接，该内部总线可以是ISA(Industry Standard Architecture，工业标准体系结构)总线、PCI(Peripheral Component Interconnect，外设部件互连标准)总线或EISA(Extended Industry Standard Architecture，扩展工业标准结构)总线等。总线可以分为地址总线、数据总线、控制总线等。为便于表示，图4中仅用一个双向箭头表示，但并不表示仅有一根总线或一种类型的总线。
存储器，用于存放程序。具体地，程序可以包括程序代码，程序代码包括计算机操作指令。存储器可以包括内存和非易失性存储器，并向处理器提供指令和数据。
处理器从非易失性存储器中读取对应的计算机程序到内存中然后运行，在逻辑层面上形成路径规划装置。处理器，执行存储器所存放的程序，并具体用于执行以下操作：
根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态；根据起点和终点确定多条候选路径；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；根据预估价值确定各候选路径中的最优路径。
上述如本申请图1所示实施例揭示的路径规划装置执行的方法可以应用于处理器中，或者由处理器实现。处理器可能是一种集成电路芯片，具有信号的处理能力。在实现过程中，上述方法的各步骤可以通过处理器中的硬件的集成逻辑电路或者软件形式的指令完成。上述的处理器可以是通用处理器，包括中央处理器(Central Processing Unit，CPU)、网络处理器(Network Processor，NP)等；还可以是数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field－Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体 管逻辑器件、分立硬件组件。可以实现或者执行本申请实施例中的公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。结合本申请实施例所公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器，处理器读取存储器中的信息，结合其硬件完成上述方法的步骤。
该电子设备还可执行图1中路径规划装置执行的方法，并实现路径规划装置在图1所示实施例的功能。本实施例中电子设备执行的其他具体功能可以参见上述方法实施例的相关内容，在此不再赘述。
本申请实施例还提出了一种计算机可读存储介质，该计算机可读存储介质存储一个或多个程序，该一个或多个程序包括指令，该指令当被包括多个应用程序的电子设备执行时，能够使该电子设备执行图1所示实施例中路径规划装置执行的方法，并具体用于执行：
根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态；根据起点和终点确定多条候选路径；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；根据预估价值确定各候选路径中的最优路径。
本实施例中计算机可读存储介质存储的程序执行的其他具体功能可以参见上述方法实施例的相关内容，在此不再赘述。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
以上所述仅为本申请的实施例而已，并不用于限制本申请。对于本领域技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (12)

  1. 一种路径规划方法，其中，所述方法包括：
    根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，所述动态路况信息包括路网中多个智能体的运动状态；
    根据起点和终点确定多条候选路径；
    利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；
    将所述环境特征输入到所述路径规划模型的价值估算网络，得到所述价值估算网络输出的各候选路径的预估价值；
    根据所述预估价值确定各候选路径中的最优路径。
  2. 如权利要求1所述的方法，其中，所述利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征包括：
    对候选路径中包含的路段，从所述环境模型中提取路段的静态环境特征和动态环境特征，将路段的静态环境特征和动态环境特征进行拼接，得到路段环境特征；
    利用长短期记忆神经网络LSTM，根据候选路径中各路段的路段环境特征，确定候选路径的环境特征。
  3. 如权利要求2所述的方法，其中，所述静态路网信息包括路段的矢量表示，所述根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型包括：
    将静态路网信息构建成路网有向图G＝(V，E，W)，其中V中的元素表征路网中的各路段，E中的元素表征各路段之间的关系，W中的元素表征各路段的长度；
    所述从所述环境模型中提取路段的静态环境特征和动态环境特征包括：
    利用图神经网络Graph-Bert，从所述有向图中提取特征向量作为所述路段的静态环境特征。
  4. 如权利要求2所述的方法，其中，所述根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型包括：
    根据所述动态路况信息确定各路段的动态环境属性，所述动态环境属性包括智能体数量维度和/或速度维度；
    所述利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征包括：
    利用动态环境属性的各维度的值拼接成特征向量作为动态环境特征，或者，利用神经网络从动态环境属性中提取特征向量作为动态环境特征。
  5. 如权利要求4所述的方法，其中，
    若所述智能体数量维度的值不为0，则所述速度维度的值为智能体平均速度；
    若所述智能体数量维度的值为0，则所述速度维度的值为预设的最大速度。
  6. 如权利要求1所述的方法，其中，所述预估价值为预测通行时间，该方法还包括：
    将各候选路径作为训练样本，将候选路径的预测通行时间作为样本预测值，获取候选路径的模拟通行时间和/或实际通行时间作为样本真实值；
    根据所述样本预测值和所述样本真实值确定训练损失值；
    根据所述训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新。
  7. 如权利要求6所述的方法，其中，所述根据所述样本预测值和所述样本真实值确定训练损失值包括：
    利用均方差函数计算训练损失值
    Figure PCTCN2021121113-appb-100001
    其中y为样本真实值，
    Figure PCTCN2021121113-appb-100002
    为样本预测值，m为样本数量。
  8. 如权利要求6所述的方法，其中，所述将各候选路径作为训练样本包括：
    将训练样本放入样本经验池中；
    在所述样本经验池中的训练样本数量达到预设值的情况下，执行根据所述 训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新的步骤，并在参数更新后，对所述样本经验池进行初始化。
  9. 一种路径规划装置，其中，所述装置包括：
    环境建模单元，用于根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，所述动态路况信息包括路网中多个智能体的运动状态；
    候选路径单元，用于根据起点和终点确定多条候选路径；
    特征提取单元，用于利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；
    价值估算单元，用于将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；
    路径确定单元，用于根据预估价值确定各候选路径中的最优路径。
  10. 如权利要求9所述的装置，其中，
    特征提取单元，用于对候选路径中包含的路段，从环境模型中提取路段的静态环境特征和动态环境特征，将路段的静态环境特征和动态环境特征进行拼接，得到路段环境特征；利用长短期记忆神经网络LSTM，根据候选路径中各路段的路段环境特征，确定候选路径的环境特征。
  11. 如权利要求9所述的装置，其中，还包括训练单元，用于将各候选路径作为训练样本，将候选路径的预测通行时间作为样本预测值，获取候选路径的模拟通行时间和/或实际通行时间作为样本真实值；根据样本预测值和样本真实值确定训练损失值；根据训练损失值和反向传播算法对特征提取网络和时间预测网络的参数进行更新。
  12. 一种电子设备，包括：
    处理器；以及
    被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使所述处理器执行下述操作：
    根据路网的静态路网信息和动态路况信息进行环境建模，得到环境模型，动态路况信息包括路网中多个智能体的运动状态；根据起点和终点确定多条候 选路径；利用路径规划模型的特征提取网络，从环境模型中提取与各候选路径对应的环境特征；将环境特征输入到路径规划模型的价值估算网络，得到价值估算网络输出的各候选路径的预估价值；根据预估价值确定各候选路径中的最优路径。
PCT/CN2021/121113 2020-11-19 2021-09-28 一种路径规划方法、装置和电子设备 Ceased WO2022105437A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/037,914 US12492910B2 (en) 2020-11-19 2021-09-28 Path planning method, apparatus and electronic device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011304527.5A CN112529254B (zh) 2020-11-19 2020-11-19 一种路径规划方法、装置和电子设备
CN202011304527.5 2020-11-19

Publications (1)

Publication Number Publication Date
WO2022105437A1 true WO2022105437A1 (zh) 2022-05-27

Family

ID=74981378

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/121113 Ceased WO2022105437A1 (zh) 2020-11-19 2021-09-28 一种路径规划方法、装置和电子设备

Country Status (3)

Country Link
US (1) US12492910B2 (zh)
CN (1) CN112529254B (zh)
WO (1) WO2022105437A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115482660A (zh) * 2022-09-02 2022-12-16 江苏中寰卫星导航通信有限公司 路径确定方法、装置、设备及存储介质
CN115660520A (zh) * 2022-12-14 2023-01-31 昆明理工大学 一种基于交通运输物流的数据分析方法及系统
CN116046008A (zh) * 2023-02-01 2023-05-02 深圳市唯特视科技有限公司 基于态势感知的路线规划方法、系统及效能评估装置
CN116090688A (zh) * 2023-04-10 2023-05-09 中国人民解放军国防科技大学 基于改进指针网络的移动目标遍历访问序列规划方法
CN116543356A (zh) * 2023-07-05 2023-08-04 青岛国际机场集团有限公司 一种轨迹确定方法、设备及介质
CN116698047A (zh) * 2023-08-07 2023-09-05 广东南方电信规划咨询设计院有限公司 勘探设备的勘探路径规划方法及系统
CN116754828A (zh) * 2023-08-21 2023-09-15 济南瑞源智能城市开发有限公司 一种智慧隧道的能耗监测方法、设备及介质

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112529254B (zh) * 2020-11-19 2022-11-25 歌尔股份有限公司 一种路径规划方法、装置和电子设备
CN113296500B (zh) * 2021-04-30 2023-04-25 浙江吉利控股集团有限公司 一种局部路径规划方法及系统
CN115345100A (zh) * 2021-05-12 2022-11-15 北京希姆计算科技有限公司 片上网络仿真模型及动态路径规划方法、装置、多核芯片
CN115586765B (zh) * 2021-06-23 2025-09-09 炬星科技（深圳）有限公司 一种移动机器人的路径规划方法、移动机器人及存储介质
CN114925651B (zh) * 2022-04-29 2026-03-17 华为技术有限公司 一种电路走线确定方法及相关设备
CN115202355B (zh) * 2022-07-21 2023-02-28 国网安徽省电力有限公司 电力杆塔作业平台的控制方法及系统
CN118119948A (zh) * 2023-03-15 2024-05-31 上海莉莉丝科技股份有限公司 寻路和寻路图神经网络的训练方法、设备、介质和计算机程序产品
EP4478304A1 (en) * 2023-06-13 2024-12-18 Zenseact AB Model-based road estimation
CN116611761B (zh) * 2023-07-18 2024-01-05 深圳市恒顺合鑫科技有限公司 一种基于智慧物流的信息处理方法及系统
CN118092413B (zh) * 2023-11-06 2025-07-22 合肥工业大学 一种多机器人系统的路径规划方法、终端及存储介质
CN117875083B (zh) * 2024-02-20 2024-09-24 清华大学 基于图论的城市多车对抗场景建模方法及装置
CN118411826B (zh) * 2024-05-27 2024-11-15 陕西四维数邦科技有限公司 一种运输车辆动态规划方法、装置、电子设备与存储介质
CN119598654B (zh) * 2024-11-20 2025-11-28 南方电网新能设计研究院(广东)有限公司 基于人工智能推理模型的电缆线路优化方法、装置、设备
CN120128680B (zh) * 2025-03-18 2025-10-24 北京博创安泰科技有限公司 面向复杂地形的超视距单兵图传中继增强装置及方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107180530A (zh) * 2017-05-22 2017-09-19 北京航空航天大学 一种基于深度时空卷积循环网络的路网状态预测方法
US20190033085A1 (en) * 2017-07-27 2019-01-31 Waymo Llc Neural Networks for Vehicle Trajectory Planning
CN110375760A (zh) * 2019-07-29 2019-10-25 北京百度网讯科技有限公司 路线确定方法、装置、设备和介质
CN110823236A (zh) * 2019-10-12 2020-02-21 百度在线网络技术（北京）有限公司 路径规划方法、装置、电子设备及存储介质
CN111489553A (zh) * 2020-04-26 2020-08-04 百度在线网络技术（北京）有限公司 一种路线规划方法、装置、设备和计算机存储介质
CN112529254A (zh) * 2020-11-19 2021-03-19 歌尔股份有限公司 一种路径规划方法、装置和电子设备

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104121918A (zh) * 2013-04-25 2014-10-29 北京搜狗信息服务有限公司 一种实时路径规划的方法和系统
CN108072381B (zh) * 2016-11-18 2020-10-27 中国移动通信有限公司研究院 一种路径规划的方法及装置
CN107944605B (zh) * 2017-11-10 2021-07-30 河海大学常州校区 一种基于数据预测的动态交通路径规划方法
CN108597246B (zh) * 2017-12-11 2020-10-13 武汉大学 一种解决路径选择实时性问题以避免局部拥堵的方法
US20190184561A1 (en) * 2017-12-15 2019-06-20 The Regents Of The University Of California Machine Learning based Fixed-Time Optimal Path Generation
CN110861634B (zh) * 2018-08-14 2023-01-17 本田技研工业株式会社 交互感知决策
CN109506669B (zh) * 2018-12-28 2021-10-08 斑马网络技术有限公司 动态路径规划方法、装置、系统以及存储介质
CN110009906B (zh) * 2019-03-25 2021-07-27 上海交通大学 基于交通预测的动态路径规划方法
CN110119839B (zh) * 2019-04-24 2021-05-14 华南理工大学 一种城市道路交通事故应急救援路径规划方法
CN110108292B (zh) * 2019-06-12 2020-03-27 山东师范大学 车辆导航路径规划方法、系统、设备及介质
CN111566583A (zh) * 2019-10-04 2020-08-21 香港应用科技研究院有限公司 自适应路径规划的系统和方法
CN111161535B (zh) * 2019-12-23 2021-03-02 山东大学 基于注意力机制的图神经网络交通流量预测方法及系统
CN111426326B (zh) * 2020-01-17 2022-03-08 深圳市镭神智能系统有限公司 一种导航方法、装置、设备、系统及存储介质
CN111368881B (zh) * 2020-02-19 2023-10-03 浙江工业大学 基于多维数据融合分析的低频gps轨迹路网匹配方法
CN111833605B (zh) * 2020-07-10 2022-04-26 北京嘀嘀无限科技发展有限公司 路况预测方法、路况预测模型训练方法、装置及存储介质
US11628858B2 (en) * 2020-09-15 2023-04-18 Baidu Usa Llc Hybrid planning system for autonomous vehicles
CN112561146B (zh) * 2020-12-08 2023-04-18 哈尔滨工程大学 一种基于模糊逻辑和深度lstm的大规模实时交通流预测方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107180530A (zh) * 2017-05-22 2017-09-19 北京航空航天大学 一种基于深度时空卷积循环网络的路网状态预测方法
US20190033085A1 (en) * 2017-07-27 2019-01-31 Waymo Llc Neural Networks for Vehicle Trajectory Planning
CN110375760A (zh) * 2019-07-29 2019-10-25 北京百度网讯科技有限公司 路线确定方法、装置、设备和介质
CN110823236A (zh) * 2019-10-12 2020-02-21 百度在线网络技术（北京）有限公司 路径规划方法、装置、电子设备及存储介质
CN111489553A (zh) * 2020-04-26 2020-08-04 百度在线网络技术（北京）有限公司 一种路线规划方法、装置、设备和计算机存储介质
CN112529254A (zh) * 2020-11-19 2021-03-19 歌尔股份有限公司 一种路径规划方法、装置和电子设备

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115482660A (zh) * 2022-09-02 2022-12-16 江苏中寰卫星导航通信有限公司 路径确定方法、装置、设备及存储介质
CN115660520A (zh) * 2022-12-14 2023-01-31 昆明理工大学 一种基于交通运输物流的数据分析方法及系统
CN115660520B (zh) * 2022-12-14 2023-03-28 昆明理工大学 一种基于交通运输物流的数据分析方法及系统
CN116046008A (zh) * 2023-02-01 2023-05-02 深圳市唯特视科技有限公司 基于态势感知的路线规划方法、系统及效能评估装置
CN116090688A (zh) * 2023-04-10 2023-05-09 中国人民解放军国防科技大学 基于改进指针网络的移动目标遍历访问序列规划方法
CN116543356A (zh) * 2023-07-05 2023-08-04 青岛国际机场集团有限公司 一种轨迹确定方法、设备及介质
CN116543356B (zh) * 2023-07-05 2023-10-27 青岛国际机场集团有限公司 一种轨迹确定方法、设备及介质
CN116698047A (zh) * 2023-08-07 2023-09-05 广东南方电信规划咨询设计院有限公司 勘探设备的勘探路径规划方法及系统
CN116698047B (zh) * 2023-08-07 2023-11-17 广东南方电信规划咨询设计院有限公司 勘探设备的勘探路径规划方法及系统
CN116754828A (zh) * 2023-08-21 2023-09-15 济南瑞源智能城市开发有限公司 一种智慧隧道的能耗监测方法、设备及介质
CN116754828B (zh) * 2023-08-21 2023-12-01 济南瑞源智能城市开发有限公司 一种智慧隧道的能耗监测方法、设备及介质

Also Published As

Publication number Publication date
US20230408278A1 (en) 2023-12-21
US12492910B2 (en) 2025-12-09
CN112529254B (zh) 2022-11-25
CN112529254A (zh) 2021-03-19

Similar Documents

Publication Publication Date Title
CN112529254B (zh) 一种路径规划方法、装置和电子设备
KR102001798B1 (ko) 비디오 이해를 위한 모션-기반 어텐션에 의한 순환 네트워크들
CN115062202B (zh) 驾驶行为意图及轨迹的预测方法、装置、设备及存储介质
CN114084155A (zh) 预测型智能汽车决策控制方法、装置、车辆及存储介质
CN117473032B (zh) 基于一致扩散的场景级多智能体轨迹生成方法及装置
CN113191550A (zh) 地图匹配方法及装置
CN109213134A (zh) 生成自动驾驶策略的方法和装置
CN115112141A (zh) 一种车辆路径规划方法、系统、电子设备及存储介质
KR20200046189A (ko) 생성적 적대 신경망에 기반한 협업 필터링을 위한 방법 및 시스템
CN113643532A (zh) 一种区域交通的预测方法以及设备
CN113849400A (zh) 一种测试场景库生成方法、装置、设备及存储介质
CN115862416A (zh) 一种路径规划方法、装置、电子设备及存储介质
CN117236441A (zh) 预测标志物与车道的关联关系的方法、装置及电子设备
CN116737540A (zh) 面向自动驾驶测试的多样关键场景生成方法、装置及设备
CN116797231A (zh) 一种风险识别模型的训练方法及装置
CN112965500B (zh) 一种过必经点集且有额外硬约束的路径规划方法及设备
CN114964293A (zh) 车辆的路径规划方法、装置及电子设备、存储介质
CN115946719A (zh) 车辆决策方法、装置及电子设备、存储介质
CN118397037A (zh) 运动目标的轨迹预测方法及装置
CN115903794A (zh) 一种多边形机器人的路径生成方法及设备
CN114445577A (zh) 一种基于图网络的预计达到时间计算方法和系统
CN114861017A (zh) 兴趣点推荐方法、装置、计算机设备和存储介质
Feng et al. Learning-based gaussian belief propagation for bundle adjustment in visual slam
Arasteh Network-aware multi-agent reinforcement learning for adaptive navigation of vehicles in a dynamic road network
Mekni Hierarchical path planning for situated agents in informed virtual geographic environments.

Legal Events

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

Ref document number: 21893588

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18037914

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21893588

Country of ref document: EP

Kind code of ref document: A1