Autonomous vehicle and trajectory planning and model training method, device and equipment

Abstract

The present disclosure provides an automatic driving vehicle and a trajectory planning and model training method, device and equipment, relates to the technical field of artificial intelligence, and in particular to the technical field of automatic driving, planning control and the like. The training method of the automatic driving trajectory planning model comprises an encoder and a planner, and the method comprises: acquiring training data, wherein the training data comprises scene information samples of an automatic driving vehicle; performing encoding processing on the scene information samples by using the encoder to obtain initial embedding vectors; performing updating processing on the initial embedding vectors to obtain target embedding vectors; performing planning processing on the target embedding vectors by using the planner to obtain a predicted trajectory; constructing a loss function based on the predicted trajectory; and adjusting model parameters of the encoder and model parameters of the planner by using the loss function. The present disclosure can improve the generalization of trajectory planning.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, specifically to the fields of autonomous driving, decision planning and other technologies, and in particular to an autonomous vehicle and a method, apparatus and device for trajectory planning and model training. Background Technology

[0002] Autonomous vehicles, also known as driverless vehicles, computer-driven vehicles, or wheeled mobile robots, are intelligent vehicles that achieve driverless operation through computer systems.

[0003] One of the planning tasks for autonomous vehicles is trajectory planning. The purpose of trajectory planning is to calculate a collision-free, executable trajectory to ensure that the vehicle can drive safely from the starting point to the destination as efficiently as possible. Summary of the Invention

[0004] This disclosure provides an autonomous vehicle and a method, apparatus, and device for trajectory planning and model training.

[0005] According to one aspect of this disclosure, a training method for an autonomous driving trajectory planning model is provided. The model includes an encoder and a planner. The method includes: acquiring training data, the training data including scene information samples of an autonomous vehicle; encoding the scene information samples using the encoder to obtain an initial embedding vector; updating the initial embedding vector to obtain a target embedding vector; planning the target embedding vector using the planner to obtain a predicted trajectory; constructing a loss function based on the predicted trajectory; and adjusting the model parameters of the encoder and the planner using the loss function.

[0006] According to another aspect of this disclosure, an autonomous driving trajectory planning method is provided, comprising: using an encoder employing an autonomous driving trajectory planning model to process scene information of an autonomous vehicle to obtain an embedding vector; using a planner employing the autonomous driving trajectory planning model to process the embedding vector to generate a target trajectory; wherein the encoder and the planner are trained using the method described in any of the preceding aspects.

[0007] According to another aspect of this disclosure, a training apparatus for an autonomous driving trajectory planning model is provided. The model includes an encoder and a planner. The apparatus includes: an acquisition module for acquiring training data, the training data including scene information samples of an autonomous vehicle; an encoding module for encoding the scene information samples using the encoder to obtain an initial embedding vector; an update module for updating the initial embedding vector to obtain a target embedding vector; a generation module for planning the target embedding vector using the planner to obtain a predicted trajectory; a construction module for constructing a loss function based on the predicted trajectory; and an adjustment module for adjusting the model parameters of the encoder and the model parameters of the planner using the loss function.

[0008] According to another aspect of this disclosure, an autonomous driving trajectory planning device is provided, comprising: an encoding module for processing scene information of an autonomous vehicle using an encoder employing an autonomous driving trajectory planning model to obtain an embedding vector; and a planning module for processing the embedding vector using a planner employing the autonomous driving trajectory planning model to generate a target trajectory; wherein the encoder and the planner are trained using the method described in any of the preceding aspects.

[0009] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to said at least one processor; wherein the memory stores instructions executable by said at least one processor, said instructions being executed by said at least one processor to enable said at least one processor to perform the method as described in any of the foregoing aspects.

[0010] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method according to any of the preceding aspects.

[0011] According to another aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method according to any of the preceding aspects.

[0012] According to another aspect of this disclosure, an autonomous vehicle is provided, including the electronic equipment described in any of the foregoing aspects.

[0013] According to the technical solution disclosed herein, the generalizability of trajectory planning can be improved.

[0014] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0015] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0016] Figure 1 This is a schematic diagram based on the first embodiment of the present disclosure;

[0017] Figure 2 This is a schematic diagram of an autonomous driving system provided according to an embodiment of the present disclosure;

[0018] Figure 3 This is a schematic diagram of an autonomous driving trajectory planning model provided according to an embodiment of this disclosure;

[0019] Figure 4 This is a schematic diagram of the training process of an autonomous driving trajectory planning model provided according to an embodiment of this disclosure;

[0020] Figure 5 This is a schematic diagram according to the second embodiment of the present disclosure;

[0021] Figure 6 This is a schematic diagram according to the third embodiment of the present disclosure;

[0022] Figure 7 This is a schematic diagram according to the fourth embodiment of the present disclosure;

[0023] Figure 8 This is a schematic diagram according to the fifth embodiment of the present disclosure;

[0024] Figure 9 This is a schematic diagram according to the sixth embodiment of the present disclosure;

[0025] Figure 10 This is a schematic diagram of an electronic device used to implement the autonomous driving trajectory planning method or the training method of the autonomous driving trajectory planning model according to the embodiments of this disclosure. Detailed Implementation

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] In related technologies, trajectory planning is typically based on preset rules. For example, a series of rules are formulated for large vehicles, blind spots, and narrow intersections. These rules can record trajectory planning schemes for corresponding situations to optimize traffic efficiency. However, rule-based approaches have problems with generalization and other aspects.

[0028] To improve the effectiveness of autonomous driving trajectory planning, the present disclosure provides the following embodiments.

[0029] Figure 1 Based on a schematic diagram of the first embodiment of this disclosure, this embodiment provides a method for training an autonomous driving trajectory planning model. The model includes an encoder and a planner. The method includes:

[0030] 101. Obtain training data, which includes scene information samples of autonomous vehicles.

[0031] 102. The scene information sample is encoded using the encoder to obtain an initial embedding vector.

[0032] 103. Update the initial embedding vector to obtain the target embedding vector.

[0033] 104. The planner is used to perform planning processing on the target embedding vector to obtain the predicted trajectory.

[0034] 105. Construct a loss function based on the predicted trajectory.

[0035] 106. Adjust the model parameters of the encoder and the model parameters of the planner using the loss function.

[0036] Scene information samples can be obtained from publicly available sample sets.

[0037] Scene information samples may include: static and / or dynamic information within a preset historical time (e.g., 6 seconds of history). Static information may include, for example, map information, and dynamic information may include, for example, the position and speed of obstacles.

[0038] After obtaining scene information samples, the scene information samples are input into the encoder. The encoder is used to convert the scene information samples into a vector, which is called the initial embedding vector.

[0039] To improve traffic efficiency and reduce traffic risks, the initial embedding vector can be updated to obtain the target embedding vector.

[0040] After obtaining the target embedding vector, the target embedding vector is input into the planner, and the planner outputs the predicted trajectory.

[0041] After obtaining the predicted trajectory, a loss function can be constructed using supervised or self-supervised methods. Taking the supervised method as an example, after obtaining scene information samples, the corresponding real trajectories can be obtained through manual annotation or other methods. When constructing the loss function, it can be based on the predicted trajectory and the real trajectory. The loss function can be, for example, the minimum mean square error (MSE) loss function.

[0042] After constructing the loss function, the model parameters can be adjusted using methods such as back propagation (BP) until a preset termination condition is met, resulting in the final autonomous driving trajectory planning model (referred to as the model). The termination condition may include reaching a preset number of iterations or meeting a preset model convergence condition. Subsequently, this model is used for trajectory planning during the inference phase.

[0043] The encoder and planner mentioned above are deep learning models, specifically the encoder and decoder in the Transformer model, or the encoder and decoder in the Long Short-Term Memory (LSTM) network model.

[0044] In this embodiment, by training an autonomous driving trajectory planning model, trajectory planning can be performed using this model, which can improve generalization compared to rule-based methods. In addition, by updating the initial embedding vector to obtain the target embedding vector, trajectory prediction can be performed based on the target embedding vector, thereby improving the accuracy of the predicted trajectory and enhancing the trajectory prediction effect.

[0045] To better understand the embodiments of this disclosure, the application scenarios of these embodiments are described. These embodiments can be applied to autonomous driving scenarios.

[0046] Autonomous driving functions in self-driving vehicles are achieved through autonomous driving systems. Currently, the levels of autonomous driving implemented by these systems can be categorized from L0 to L5. L0 represents no autonomous driving, i.e., traditional manual driving by a driver. L1, also known as driver assistance, includes basic functions such as cruise control, automatic parking, and lane keeping. L2, also known as semi-autonomous driving, includes functions such as automatic assisted driving and hazard prediction braking. L3, also known as conditional autonomous driving, achieves fully automated driving under normal road conditions, but manual assistance is still required in emergency situations. L4 represents highly automated driving, where the vehicle's overall braking performance and responsiveness have reached a high level; the driver does not need to manually control the vehicle, and the ride is smooth and stable. L5 represents unconditional fully autonomous driving, eliminating concerns about road conditions and weather conditions under any circumstances.

[0047] like Figure 2 As shown, the core modules of the autonomous driving system 200 include: a high-definition map (HD map) 201, a localization system 202, a perception system 203, a prediction system 204, a global navigation system 205, a planning system 206, and a control module 207.

[0048] Among them, High-Precision Map 201, also known as Autonomous Driving Map or High-Resolution Map, is a new map data paradigm for autonomous vehicles. High-precision maps have an absolute positional accuracy close to 1 meter and a relative positional accuracy at the centimeter level, reaching 10-20 cm.

[0049] Positioning system 202 can provide high-precision (centimeter-level) positioning services based on positioning devices and high-precision maps. Positioning devices include, for example, one or more of the Global Positioning System (GPS), Global Navigation Satellite System (GNSS), and Inertial Navigation System (INS).

[0050] The perception system 203 provides comprehensive environmental perception services for autonomous vehicles. Specifically, it may include devices such as cameras, LiDAR, millimeter-wave radar, and ultrasonic radar.

[0051] The prediction system 204, taking data from the perception system as input, extracts historical motion parameters of the vehicle and / or obstacles, and combines them with Kalman filtering, neural networks, and other techniques to infer the future motion trajectory of the vehicle and / or obstacles. The predicted motion trajectory can be provided to the planning system.

[0052] The global navigation system 205 is used to obtain the optimal global navigation path that meets the performance evaluation indicators based on the vehicle's initial position and target position, combined with the road network topology, through a global path search algorithm.

[0053] Planning System 206 primarily provides services such as vehicle obstacle avoidance, lane change decision-making, path planning, trajectory planning, and speed planning.

[0054] The control system 207 is used to perform longitudinal and lateral tracking control based on the driving trajectory provided by the decision-making and planning system. Specifically, it can control the vehicle's chassis system to perform steering, throttle, braking, and other operations via the Controller Area Network (CAN) bus.

[0055] In the scenario of autonomous driving trajectory planning, the autonomous driving trajectory planning method can be executed by the planning system of the autonomous vehicle.

[0056] In related technologies, trajectory planning can be performed based on preset rules. That is, the correspondence between driving state and trajectory planning can be pre-configured, and the trajectory planning result can be determined based on the correspondence. However, since it is difficult to exhaust all driving states, there is a problem of poor generalization.

[0057] To improve generalization, this embodiment uses machine learning for trajectory planning. The machine learning model used for trajectory planning can be called an autonomous driving trajectory planning model, such as... Figure 3 As shown, the model includes an encoder 301 and a planner 302.

[0058] During the operation of an autonomous vehicle, the autonomous driving system can acquire scene information within a preset historical time period (e.g., 6 seconds). Scene information includes, for example, map information provided by a map and positioning system, and / or dynamic information such as the position and speed of obstacles provided by a perception system. Then, the scene information is input into encoder 301, which encodes the scene information to obtain an embedding vector. The embedding vector is input into planner 302, which decodes the embedding vector to generate a target trajectory, which is the planned trajectory.

[0059] To improve the traffic efficiency of the target trajectory generated by the autonomous driving trajectory planning model, a predictor can be introduced to train the autonomous driving trajectory planning model during the training phase.

[0060] like Figure 4 As shown, during the training phase, in addition to the encoder 401 and planner 402 included in the autonomous driving trajectory planning model, a predictor 403 is also included. The encoder can be an encoder of a Transformer network or an LSTM network, and the planner and predictor can be decoders of a Transformer network or an LSTM network.

[0061] Encoder 401 is used to encode scene information samples to obtain an initial embedding vector;

[0062] Predictor 403 is used to obtain the passage parameters based on the initial embedding vector.

[0063] Traffic parameters may include traffic efficiency and / or traffic risk. This embodiment takes traffic efficiency and traffic risk as examples. Furthermore, traffic efficiency and traffic risk can be obtained using the same predictor, or they can be obtained using separate predictors.

[0064] After obtaining the access parameters, the initial embedding vector can be updated using the access parameters to obtain the target embedding vector. The embedding vector can be represented by z.

[0065] Specifically, for traffic efficiency, a gradient ascent method is used for updating to obtain higher traffic efficiency, while for traffic risk, a gradient descent method is used for updating to obtain lower traffic risk.

[0066] Planner 402 is used to obtain the predicted trajectory based on the target embedding vector.

[0067] Taking supervised training as an example, a loss function can be constructed based on the predicted trajectory and the real trajectory corresponding to the scene information sample. Then, the model parameters of the encoder, the model parameters of the planner, and the model parameters of the predictor can be adjusted using the loss function.

[0068] In conjunction with the above application scenarios, this disclosure also provides a training method for an autonomous driving trajectory planning model.

[0069] Figure 5 Based on a schematic diagram of the second embodiment of this disclosure, this embodiment provides a method for training an autonomous driving trajectory planning model. The model includes an encoder and a planner. The method includes:

[0070] 501. Obtain training data, wherein the training data includes: scene information samples of autonomous vehicles.

[0071] 502. The scene information sample is encoded using the encoder to obtain an initial embedding vector.

[0072] 503. A predictor is used to perform prediction processing on the initial embedding vector to obtain the passage parameters.

[0073] 504. Update the initial embedding vector based on the access parameters to obtain the target embedding vector.

[0074] In this embodiment, the initial embedding vector is updated based on the passage parameters to obtain the target embedding vector. This allows the target embedding vector to contain passage information, thereby enabling the acquisition of a target trajectory with better passage performance, improving the passage efficiency of the target trajectory and reducing passage risks.

[0075] Taking the traffic parameters, which include traffic efficiency and traffic risk, as an example, the traffic efficiency is updated using a gradient ascent method, and the traffic risk is updated using a gradient descent method.

[0076] Specifically, the passage parameters include: passage efficiency, and / or, passage risk;

[0077] The step of updating the initial embedding vector based on the passage parameters to obtain the target embedding vector includes:

[0078] Determine the first gradient of the traffic efficiency;

[0079] Determine the second gradient of the passage risk;

[0080] The initial embedding vector is updated based on the product of the first weight and the first gradient, and / or the product of the second weight and the second gradient, to obtain the target embedding vector; the first weight is a positive value and the second weight is a negative value.

[0081] Specifically, the calculation formula can be:

[0082]

[0083] Where z' is the target embedding vector; z is the initial embedding vector; f efficency It refers to traffic efficiency; f risk It represents the risk of passage; α1 is the first weight, -α2 is the second weight, and both α1 and α2 are positive values.

[0084] Since the first weight corresponding to traffic efficiency is positive, it can be considered that the traffic efficiency is optimized by gradient ascent; since the second weight corresponding to traffic risk is negative, it can be considered that the traffic risk is optimized by gradient descent.

[0085] In this embodiment, since the first weight is positive, higher traffic efficiency can be obtained, and since the second weight is negative, lower traffic risk can be obtained. Thus, through the above-mentioned update method, traffic efficiency can be improved and traffic risk can be reduced.

[0086] 505. The planner is used to perform planning processing on the target embedding vector to obtain the predicted trajectory.

[0087] 506. Construct a loss function based on the predicted trajectory.

[0088] 507. Adjust the model parameters of the encoder, the model parameters of the planner, and the model parameters of the predictor using the loss function.

[0089] The predicted trajectory can be a predicted trajectory within a preset future duration (e.g., 8 seconds), and can be represented as ((x1,y1),(x2,y2),...(x n ,y n )) indicates that, correspondingly, traffic efficiency may include: the distance traveled within the future time period, and traffic risk may include: the probability of a risk occurring within the future time period.

[0090] In this embodiment, by setting the above-mentioned traffic efficiency and traffic risk, traffic efficiency and traffic risk can be obtained simply and efficiently.

[0091] Furthermore, the probability of a risk occurring in the future may include at least one of the following: probability of collision, probability of emergency braking, probability of takeover.

[0092] In this embodiment, by acquiring the aforementioned traffic risks, the comprehensiveness and accuracy of traffic risks can be improved.

[0093] Figure 6 Based on a schematic diagram of the third embodiment of this disclosure, this embodiment provides an autonomous driving trajectory planning method, the method comprising:

[0094] 601. An encoder using an autonomous driving trajectory planning model processes scene information of autonomous vehicles to obtain embedding vectors.

[0095] 602. The planner of the autonomous driving trajectory planning model is used to process the embedded vector to generate the target trajectory.

[0096] The encoder and the planner are trained using the training method shown in any of the above embodiments.

[0097] In this embodiment, the target trajectory is generated using a model, which can improve generalization compared to the rule-based approach.

[0098] Figure 7 The diagram is based on the fourth embodiment of the present disclosure. This embodiment provides a training device for an autonomous driving trajectory planning model. The model includes an encoder and a planner. The device 700 includes an acquisition module 701, an encoding module 702, an update module 703, a generation module 704, a construction module 705, and an adjustment module 706.

[0099] The acquisition module 701 is used to acquire training data, which includes scene information samples of autonomous vehicles; the encoding module 702 is used to encode the scene information samples using the encoder to obtain an initial embedding vector; the update module 703 is used to update the initial embedding vector to obtain a target embedding vector; the generation module 704 is used to plan the target embedding vector using the planner to obtain a predicted trajectory; the construction module 705 is used to construct a loss function based on the predicted trajectory; and the adjustment module 706 is used to adjust the model parameters of the encoder and the model parameters of the planner using the loss function.

[0100] In this embodiment, by training an autonomous driving trajectory planning model, trajectory planning can be performed using this model, which can improve generalization compared to rule-based methods. In addition, by updating the initial embedding vector to obtain the target embedding vector, trajectory prediction can be performed based on the target embedding vector, thereby improving the accuracy of the predicted trajectory and enhancing the trajectory prediction effect.

[0101] In some embodiments, the update module 703 is further configured to:

[0102] The initial embedding vector is processed by a predictor to obtain the passage parameters;

[0103] The initial embedding vector is updated based on the access parameters to obtain the target embedding vector.

[0104] In this embodiment, the initial embedding vector is updated based on the passage parameters to obtain the target embedding vector. This allows the target embedding vector to contain passage information, thereby enabling the acquisition of a target trajectory with better passage performance, improving the passage efficiency of the target trajectory and reducing passage risks.

[0105] In some embodiments, the passage parameters include: passage efficiency, and / or, passage risk;

[0106] The update module 703 is further used for:

[0107] Determine the first gradient of the traffic efficiency;

[0108] Determine the second gradient of the passage risk;

[0109] The initial embedding vector is updated based on the product of the first weight and the first gradient, and / or the product of the second weight and the second gradient, to obtain the target embedding vector; the first weight is a positive value and the second weight is a negative value.

[0110] In this embodiment, since the first weight is positive, higher traffic efficiency can be obtained, and since the second weight is negative, lower traffic risk can be obtained. Thus, through the above-mentioned update method, traffic efficiency can be improved and traffic risk can be reduced.

[0111] In some embodiments, the predicted trajectory includes: a predicted trajectory for a preset future duration;

[0112] The traffic efficiency includes: the travel distance over the future time period;

[0113] The passage risk includes the probability of the risk occurring within the future timeframe.

[0114] In this embodiment, by setting the above-mentioned traffic efficiency and traffic risk, traffic efficiency and traffic risk can be obtained simply and efficiently.

[0115] In some embodiments, the probability of a risk occurring within the future timeframe includes at least one of the following: probability of collision, probability of emergency braking, and probability of takeover.

[0116] In this embodiment, by acquiring the aforementioned traffic risks, the comprehensiveness and accuracy of traffic risks can be improved.

[0117] Figure 8 This is a schematic diagram according to the fifth embodiment of the present disclosure. This embodiment provides an autonomous driving trajectory planning device, which includes an encoding module 801 and a planning module 802.

[0118] The encoding module 801 is used to process the scene information of the autonomous vehicle using an encoder based on the autonomous driving trajectory planning model to obtain an embedding vector; the planning module 802 is used to process the embedding vector using the planner based on the autonomous driving trajectory planning model to generate a target trajectory.

[0119] The encoder and the planner are trained using the training method shown in any of the above embodiments.

[0120] In this embodiment, the target trajectory is generated by a model, which can improve the generalization compared to the rule-based approach. In addition, by updating the initial embedding vector to obtain the target embedding vector, the accuracy of the predicted trajectory can be obtained by performing trajectory prediction based on the target embedding vector, thus improving the trajectory prediction effect.

[0121] It is understood that the same or similar content in different embodiments of this disclosure can be referred to each other.

[0122] It is understood that the terms "first" and "second" in the embodiments of this disclosure are only used for distinction and do not indicate the degree of importance or the order of events.

[0123] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0124] According to embodiments of this disclosure, such as Figure 9 As shown, this disclosure also provides an autonomous driving vehicle 900, which includes an electronic device 901. A description of the electronic device 901 can be found in subsequent embodiments. Specifically, the electronic device 901 can primarily employ an autonomous driving trajectory planning model for trajectory planning.

[0125] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0126] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement embodiments of the present disclosure is shown. The electronic device 1000 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, servers, blade servers, mainframe computers, and other suitable computers. The electronic device 1000 may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0127] like Figure 10 As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of the electronic device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0128] Multiple components in electronic device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of displays, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows electronic device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0129] The computing unit 1001 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1001 performs the various methods and processes described above, such as autonomous driving trajectory planning methods or autonomous driving trajectory planning model training methods. For example, in some embodiments, the autonomous driving trajectory planning method or autonomous driving trajectory planning model training method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the autonomous driving trajectory planning method or autonomous driving trajectory planning model training method described above can be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured by any other suitable means (e.g., by means of firmware) to execute an autonomous driving trajectory planning method or an autonomous driving trajectory planning model training method.

[0130] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0131] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to the processor or controller of a general-purpose computer, special-purpose computer, or other programmable load balancing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0132] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0135] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0136] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A training method for an autonomous driving trajectory planning model, the model comprising: The encoder and planner, the method comprising: Acquire training data, which includes: scene information samples of autonomous vehicles; The scene information samples are encoded using the encoder to obtain an initial embedding vector; Updating the initial embedding vector to obtain the target embedding vector includes: using a predictor to predict the initial embedding vector to obtain passage parameters, the passage parameters including: passage efficiency, and / or, passage risk; updating the initial embedding vector based on the gradient of the passage parameters to obtain the target embedding vector. The planner is used to perform planning processing on the target embedding vector to obtain the predicted trajectory; A loss function is constructed based on the predicted trajectory; The model parameters of the encoder and the model parameters of the planner are adjusted using the loss function.

2. The method according to claim 1, wherein, The step of updating the initial embedding vector based on the gradient of the passage parameters to obtain the target embedding vector includes: Determine the first gradient of the traffic efficiency; Determine the second gradient of the passage risk; The initial embedding vector is updated based on the product of the first weight and the first gradient, and / or the product of the second weight and the second gradient, to obtain the target embedding vector; the first weight is a positive value and the second weight is a negative value.

3. The method according to claim 1, wherein, The predicted trajectory includes: a predicted trajectory for a preset future duration; The traffic efficiency includes: the travel distance over the future time period; and / or... The passage risk includes the probability of the risk occurring within the future timeframe.

4. The method according to claim 3, wherein, The probability of the risk occurring within the future timeframe includes at least one of the following: Collision probability, emergency braking probability, and takeover probability.

5. An autonomous driving trajectory planning method, comprising: An encoder employing an autonomous driving trajectory planning model processes scene information from autonomous vehicles to obtain embedding vectors. The planner of the aforementioned autonomous driving trajectory planning model processes the embedded vector to generate the target trajectory; The encoder and the planner are trained using the method described in any one of claims 1-4.

6. A training device for an autonomous driving trajectory planning model, the model comprising: The encoder and planner, the device comprising: An acquisition module is used to acquire training data, which includes scene information samples of autonomous vehicles; The encoding module is used to encode the scene information samples using the encoder to obtain an initial embedding vector; An update module is used to update the initial embedding vector to obtain a target embedding vector, including: using a predictor to predict the initial embedding vector to obtain passage parameters, the passage parameters including: passage efficiency, and / or, passage risk; updating the initial embedding vector based on the gradient of the passage parameters to obtain the target embedding vector; A generation module is used to perform planning processing on the target embedding vector using the planner to obtain a predicted trajectory; A construction module is used to construct a loss function based on the predicted trajectory; An adjustment module is used to adjust the model parameters of the encoder and the model parameters of the planner using the loss function.

7. The apparatus according to claim 6, wherein, The update module is further used for: Determine the first gradient of the traffic efficiency; Determine the second gradient of the passage risk; The initial embedding vector is updated based on the product of the first weight and the first gradient, and / or the product of the second weight and the second gradient, to obtain the target embedding vector. The first weight is a positive value, and the second weight is a negative value.

8. The apparatus according to claim 6, wherein, The predicted trajectory includes: a predicted trajectory for a preset future duration; The traffic efficiency includes: the travel distance over the future time period; and / or... The passage risk includes the probability of the risk occurring within the future timeframe.

9. The apparatus according to claim 8, wherein, The probability of the risk occurring within the future timeframe includes at least one of the following: Collision probability, emergency braking probability, and takeover probability.

10. An autonomous driving trajectory planning device, comprising: The encoding module is used to process scene information of autonomous vehicles using an encoder based on an autonomous driving trajectory planning model to obtain embedding vectors. The planning module is used by the planner of the autonomous driving trajectory planning model to process the embedded vector to generate the target trajectory; The encoder and the planner are trained using the method described in any one of claims 1-4.

11. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

12. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

13. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-5.

14. An autonomous vehicle, comprising: The electronic device as claimed in claim 11.

Images