An autonomous driving multi-agent future behavior topology inference method, device, equipment, medium and product

Abstract

The present invention discloses a method, apparatus, device, medium and product for inferring the future behavior topology of multiple autonomous agents. The method includes: obtaining the position information of a current vehicle and the position information of each agent in an agent set included in the scene where the current vehicle is located; inputting the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology inference result, wherein the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology inference result sample corresponding to the agent set sample. Through the technical solution of the present invention, the consistency and accuracy of jointly predicting and planning the behaviors of multiple agents can be enhanced.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of autonomous driving, and in particular, to a method, device, equipment, medium and product for inferring the future behavior topology of multiple autonomous driving agents. Background Art

[0002] The autonomous driving system aims to achieve safe and socially compliant driving through the integration of behaviors among interactive agents. However, the uncertainty and heterogeneous interactions in multi-agent driving scenarios remain the biggest challenges currently faced. To establish a more compatible future behavior representation for multiple agents, the current mainstream paradigms are divided into two types, namely dense representation and sparse representation. However, the current dense and sparse future behavior representations have problems of low efficiency and inconsistency in multi-agent modeling, resulting in unstable multi-agent behavior patterns during IPP (Integrated Prediction and Planning). Summary of the Invention

[0003] The embodiments of the present invention provide a method, device, equipment, medium and product for inferring the future behavior topology of multiple autonomous driving agents, so as to enhance the consistency and accuracy of jointly predicting and planning the behaviors of multiple agents.

[0004] According to one aspect of the present invention, there is provided a method for inferring the future behavior topology of multiple autonomous driving agents, including:

[0005] Obtaining the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located;

[0006] Inputting the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology inference result, where the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: agent set samples and future behavior topology inference result samples corresponding to the agent set samples.

[0007] According to another aspect of the present invention, there is provided a device for inferring the future behavior topology of multiple autonomous driving agents, and the device includes:

[0008] An obtaining module, configured to obtain the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located;

[0009] An input module for inputting the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology inference result, where the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology inference result sample corresponding to the agent set sample.

[0010] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0011] At least one processor; and

[0012] A memory communicatively connected to the at least one processor; wherein,

[0013] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the multi-agent future behavior topology inference method for autonomous driving according to any embodiment of the present invention.

[0014] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the multi-agent future behavior topology inference method for autonomous driving according to any embodiment of the present invention when executed.

[0015] According to another aspect of the present invention, an embodiment of the present invention further provides a computer program product, the computer program product comprising a computer program, and the computer program implements the multi-agent future behavior topology inference method for autonomous driving according to any embodiment of the present invention when executed by a processor.

[0016] In the embodiment of the present invention, a training sample set is formed by acquiring an agent set sample and a future behavior topology inference result sample corresponding to the agent set sample, an initial model is iteratively trained to obtain a collaborative learning model, the position information of the current vehicle and the position information of each agent in the agent set included in the current vehicle's scene are acquired, and the position information of the current vehicle and the position information of each agent are input into the collaborative learning model to obtain a future behavior topology inference result. Through the technical solution of the present invention, the consistency and accuracy of jointly predicting and planning the behaviors of multiple agents can be enhanced.

[0017] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Description of the Drawings

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0019] Figure 1 is a flowchart of a method for inferring the future behavior topology of multiple intelligent agents in an autonomous driving in an embodiment of the present invention;

[0020] Figure 2 is a schematic structural diagram of a device for inferring the future behavior topology of multiple intelligent agents in an autonomous driving in an embodiment of the present invention;

[0021] Figure 3 is a schematic structural diagram of an electronic device for implementing the method for inferring the future behavior topology of multiple intelligent agents in an autonomous driving in an embodiment of the present invention. Specific Embodiments

[0022] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and their derivatives are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0024] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, usage scopes, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0025] Embodiment 1

[0026] Figure 1 It is a flowchart of a method for inferring the future behavior topology of multiple autonomous agents in an embodiment of the present invention. This embodiment is applicable to the situation of inferring the future behavior topology of multiple autonomous agents. This method can be executed by the device for inferring the future behavior topology of multiple autonomous agents in the embodiment of the present invention. The device can be implemented in software and / or hardware, such as Figure 1 shown, and the method specifically includes the following steps:

[0027] S101. Obtain the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located.

[0028] In this embodiment, the current vehicle can be a self-driving vehicle, that is, a vehicle equipped with an autonomous driving system. Exemplarily, the scene where the current vehicle is located can be the driving scene passed by the current vehicle during driving, such as an intersection. It should be noted that the set of agents can be a set composed of other motor vehicles, non-motor vehicles, pedestrians, etc. other than the vehicle itself included in the driving scene passed by the current vehicle during driving. Among them, each other motor vehicle, non-motor vehicle or pedestrian is used as an agent.

[0029] Specifically, for example, in a driving scene composed of multiple agents, obtain the position information of the self-driving vehicle and the position information of other agents other than the self-driving vehicle.

[0030] S102. Input the position information of the current vehicle and the position information of each agent into the collaborative learning model to obtain the future behavior topology inference result.

[0031] In this embodiment, the collaborative learning model can be a model for predicting IPP (Integrated Prediction and Planning, joint prediction and planning) objectives, inferring the behavior topology, and generating future trajectory clusters. Preferably, in this embodiment, the collaborative learning model can be the collaborative learning model framework BeTopNet based on Transformer (BeTop, Behavioral Topology. The core of BeTop is to explicitly represent a topology supervision of the future interaction consensus of multiple agents, and establish an inference task for guiding autonomous driving prediction and planning).

[0032] Among them, the collaborative learning model is obtained by iteratively training the initial model with a training sample set. The training sample set includes: agent set samples and future behavior topology inference result samples corresponding to the agent set samples.

[0033] Among them, the initial model can be the untrained collaborative learning model framework BeTopNet based on Transformer. The intelligent agent set sample can be a set composed of each intelligent agent in the driving scenario formed by multiple pre-collected intelligent agents. The future behavior topology inference result sample corresponding to the intelligent agent set sample can be the future behavior topology inference results of each intelligent agent corresponding to the pre-determined intelligent agent set sample. These information are pre-collected as a training sample set to train the initial model to obtain a collaborative learning model.

[0034] Specifically, obtain the intelligent agent set sample and the future behavior topology inference result sample corresponding to the intelligent agent set sample to form a training sample set, and iteratively train the initial model to obtain a collaborative learning model. After the model training is completed, input the position information of the current vehicle and the position information of each intelligent agent into the trained collaborative learning model for prediction and inference to obtain the future behavior topology inference result.

[0035] In the embodiment of the present invention, by obtaining the intelligent agent set sample and the future behavior topology inference result sample corresponding to the intelligent agent set sample to form a training sample set, iteratively training the initial model to obtain a collaborative learning model, obtaining the position information of the current vehicle and the position information of each intelligent agent in the intelligent agent set included in the scene where the current vehicle is located, and inputting the position information of the current vehicle and the position information of each intelligent agent into the collaborative learning model to obtain the future behavior topology inference result. Through the technical solution of the present invention, the consistency and accuracy of jointly predicting and planning the behaviors of multiple intelligent agents can be enhanced.

[0036] Optionally, iteratively training the initial model with the training sample set includes:

[0037] Establish an initial model.

[0038] Input the intelligent agent set sample in the training sample set into the initial model to obtain the predicted future behavior topology inference result corresponding to the intelligent agent set sample.

[0039] Among them, the predicted future behavior topology inference result can be the future behavior topology inference result obtained by the initial model through prediction and inference based on the intelligent agent set sample.

[0040] Specifically, input the intelligent agent set sample in the training sample set into the initial model for prediction and inference to obtain the predicted future behavior topology inference result corresponding to the intelligent agent set sample.

[0041] Based on a preset fitness function, determine the fitness value according to the predicted future behavior topology inference result corresponding to the intelligent agent set sample.

[0042] In this embodiment, the preset fitness function may be a function derived and defined by the user according to the multi-agent behavior topology reasoning task, wherein the fitness value may be a value obtained by calculating the sample of the agent set according to the preset fitness function.

[0043] If the fitness value does not meet the preset target, the parameters of the initial model are trained according to the loss function formed by the predicted future behavior topology reasoning results corresponding to the agent set samples and the future behavior topology reasoning result samples corresponding to the agent set samples.

[0044] Preferably, in this embodiment, the preset goal may be to obtain the maximum fitness value.

[0045] Specifically, during the iterative training of the model, if the fitness value does not reach the preset target, for example, the fitness value is maximum to a certain extent, the weights and other parameters of the initial model are trained according to the loss function formed by the predicted future behavior topological reasoning results corresponding to the agent set samples and the future behavior topological reasoning result samples corresponding to the agent set samples.

[0046] Return to execute the operation of inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topological reasoning results corresponding to the agent set samples, until the fitness value meets the preset target, or the number of iterations is greater than or equal to the preset number, and the collaborative learning model is obtained.

[0047] The preset number of times may be a number threshold pre-set by the user according to actual conditions, and this embodiment does not limit this.

[0048] Specifically, the training is iterated continuously until the obtained fitness value is maximized, or the number of iterations reaches a preset number, then the parameter corresponding to the maximum fitness value is determined as the weight of the collaborative learning model, and finally a trained collaborative learning model is obtained.

[0049] Optionally, the agent set samples in the training sample set are input into the initial model to obtain the predicted future behavior topology reasoning results corresponding to the agent set samples, including:

[0050] Input the agent set samples in the training sample set into the initial model.

[0051] In actual operation, the agent set samples may include self-driving car samples and multi-agent samples such as other motor vehicles, non-motor vehicles or pedestrians.

[0052] Get the position information of each agent sample in the agent set sample.

[0053] Specifically, the position information of the self-driving car sample and the position information of multi-agent samples such as other motor vehicles, non-motor vehicles or pedestrians in the agent set sample are obtained.

[0054] Construct a topological graph based on the position information of each agent sample.

[0055] In this embodiment, the topological graph can be a behavioral topological graph for representing the interaction behavior relationship of multiple agents.

[0056] Input the topological graph into the encoder-decoder in the initial model to obtain the predicted future behavior topological inference result corresponding to the agent set sample.

[0057] In this embodiment, a collaborative learning model framework BeTopNet based on Transformer is introduced for jointly predicting the planning (IPP) objective. Among them, the synergistic decoder synergistically infers the behavioral topology through an iterative decoding module strategy and generates a future trajectory cluster. Among them, the decoding module is the encoder-decoder.

[0058] Specifically, after constructing the topological graph, input the topological graph into the encoder-decoder in the initial model to infer the behavioral topology and obtain the predicted future behavior topological inference result corresponding to the agent set sample.

[0059] Optionally, constructing a topological graph based on the position information of each agent sample includes:

[0060] Based on the topological knot theory, take each agent sample as a knot and generate a knot interweaving set according to the position information of each agent sample.

[0061] It should be noted that the knot interweaving set can be a knot group composed of all knots after taking each agent sample in the agent combination sample as a knot.

[0062] The behavioral topology BeTop is derived from the topological knot theory (braid theory), which infers the consistent interaction paradigm in multiple trajectories under the knot space transformation from the intertwine. This enables BeTop to intuitively extract the forward intertwine (occupation) as a joint topology from the multi-agent future trajectories abstracted as knot trajectories, combining dense and sparse representations. Specifically, based on the topological knot theory, take each agent sample in the agent set sample as a knot, determine the position information of each knot according to the position information of each agent sample, and generate a knot interweaving set from all knots and the corresponding position information.

[0063] Determine the topological edge set according to the knot interweaving set.

[0064] In this embodiment, every two knots can form a topological edge. There are multiple agents in the agent set sample, that is, there are multiple knots, that is, there are multiple topological edges. Among them, the topological edge set can be a set composed of all topological edges.

[0065] Among them, each topological edge represents the future interaction behavior between two agents corresponding to the topological edge.

[0066] Specifically, every two knots in the knot interweaving set form a topological edge, and finally a topological edge set is formed.

[0067] Construct a topological graph according to each knot and each topological edge.

[0068] Specifically, all knots are used as topological nodes, and all topological nodes and topological edges form a topological graph.

[0069] Optionally, the method further includes:

[0070] Obtain the knot function corresponding to each knot.

[0071] In the actual operation process, using knot theory, considering a knot group composed of multiple knots, where each knot represents a group of monotonically increasing functions, that is, a knot function.

[0072] Determine the edge topological function of the topological edge corresponding to every two knots according to every two knot functions.

[0073] In the actual operation process, each topological edge element can be defined by two knot functions, that is, the edge topological function of the topological edge formed by the two knots can be represented by the two knot functions.

[0074] Determine a preset fitness function according to the knot function corresponding to each knot and the edge topological function of each topological edge.

[0075] Exemplarily, the preset fitness function can be to obtain the maximum knot function and the maximum edge topological function.

[0076] Optionally, input the topological graph into the encoder-decoder in the initial model to obtain the predicted future behavior topological inference result corresponding to the agent set sample, including:

[0077] Input the topological graph into the encoder in the initial model to obtain the encoded scene features.

[0078] In this embodiment, the encoder can be, for example, a driving scene encoder, which is used to encode the obtained position information of the agents into the scene features of the driving scene. Among them, the encoded scene features can be the encoded driving scene features output by the encoder.

[0079] Input the encoded scene features into the decoder in the initial model to obtain the predicted future behavior topological inference result corresponding to the agent set sample.

[0080] In this embodiment, the decoder can perform collaborative inference of the behavior topology based on the encoded scene features generated by the encoder and generate future trajectory clusters.

[0081] As a detailed description of the embodiments of the present invention, the following provides a detailed introduction to the method for inferring the future behavior topology of autonomous driving multi-agents:

[0082] The autonomous driving system aims to achieve safe and socially compliant driving through the integration of behaviors among interactive agents. However, the uncertainty and heterogeneous interactions in multi-agent driving scenarios remain the biggest challenges currently faced. The current dense and sparse future behavior representations have problems of inefficiency and inconsistency in multi-agent modeling, resulting in unstable multi-agent behavior patterns during joint prediction and planning IPP. To solve this problem, this embodiment proposes a behavioral characterization paradigm of a topological structure as a compliant behavioral prospect to guide downstream trajectory generation. Specifically, this embodiment introduces Behavior Topology (BeTop), which visually represents the consensus behavior patterns among multi-agents in the future. Behavior Topology is derived from the Braid theory and is used to extract compliant interactive topologies from multi-agent future trajectories. The collaborative learning framework (BeTopNet) supervised by Behavior Topology promotes the consistency of behavior prediction and planning in the predicted topological prior. By mimicking the emergency learning paradigm, Behavior Topology (BeTop) can effectively handle the behavioral uncertainties in prediction and planning.

[0083] Multi-agent Behavior Representation: Shaping the collective future behavior of different agents is crucial for achieving socially consistent driving operations. Dense representations focus on occupancy prediction. Predictive spatial occupancy in a dense Bird's Eye View (BEV) representation provides flexibility for any agent and consistency with perception. However, fixed resolution and receptive fields lead to scene occlusion, resulting in intractable occupancy. Meanwhile, sparse representations aggregate multi-agent future behaviors into clusters of future trajectories or intentions with multi-modal. Representations for joint future behavior prediction are derived through goal-based sampling or recombination in marginal prediction. However, the set of joint modal representations is vulnerable to modal collapse and incurs exponential complexity. Topological representations have gained attention as a quantification metric for behavioral priors or scene interactions, but the topological properties describing collective future behavior have not been explored. The Behavior Topology (BeTop) of this embodiment addresses this issue by combining topologically dense behavioral probabilities with sparse motions of joint predictions to present a structured future behavior representation. On the other hand, inconsistent multi-agent interaction behaviors motivate the exploitation of future interactions. Implicit methods use attention mechanisms or Graph Neural Networks (GNNs) to obtain implicit behavior interaction information from future trajectory regression learning. However, the learning efficiency of implicit supervision is unstable in dynamic scenes. Instead, explicit reasoning of mutual behaviors through conditional probability decomposition, relational reasoning, or entropy-based methods can provide consistent behavioral priors. However, the large differences in multi-agent dynamics and driving scene geometric properties can lead to unstable reasoning. Different from this, Behavior Topology designs a compact topological supervision paradigm that can stabilize the consistent consensus interaction between multi-agent future behaviors, and it also provides topologically equivalent behavior representations to guide joint prediction and planning.

[0084] Joint Prediction and Planning: The Integrated Prediction and Planning (IPP) system aims to coordinate the trajectory learning of the future interaction behaviors among multiple agents in the ego-vehicle's driving scenario. Rule-based methods integrate manually defined future interactions to evaluate the settings of candidate plans, providing significant results in rule-driven closed-loop experiments. Nevertheless, the lack of real driving behaviors makes the rule-based paradigm show a huge gap in actual interaction scenarios. Learning-based methods generate plans that mimic human driving by integrating predictions into the overall model modeling. However, the interaction modeling based on historical inputs poses challenges to the future consistency among multiple agents. Recently, the hybrid paradigm uses the post-processing and optimization of learning-based models to achieve behavioral interaction between prediction and planning. However, this introduces a large amount of computational overhead, and the mimicking plans often overestimate the uncertainty in behavioral prediction. Works based on trees and contingency planning seek to balance the initiative and passivity of planning behaviors in the face of behavioral uncertainty. Nevertheless, paradigms that do not model the overall interaction will fall into passive planning strategies and lead to an exponential increase in prediction costs. Behavioral Topology (BeTop) provides clear priors for future interaction behaviors, thus enhancing the generation of compliant trajectories. In addition, the cooperative prediction and contingency planning model based on behavioral topology specifications effectively manages behavioral uncertainty.

[0085] Autonomous driving systems pursue driving strategies that are safe, human-like, and compliant with the driving environment. This has promoted the design of the representation of the future behaviors of multiple agents, future prediction, and strategy negotiation among multiple agents between interactive agents and autonomous vehicles (AVs). Learning-based paradigms have achieved excellent accuracy, including end-to-end modular design, scenario interaction modeling, and motion prediction and planning integration. However, due to the uncertainty of the scenario and the unstable interaction patterns of the future behaviors of multiple agents, there are huge challenges for self-driving systems in actual driving scenarios.

[0086] To establish a more compatible future behavior representation for multi-agent systems, the current mainstream paradigms are divided into two types, namely dense representation and sparse representation. The dense representation paradigm performs grid quantization on the behavior of multi-agent systems centered around self-driving vehicles, predicting the occupancy probability or temporal flow of the bird's-eye view (BEV). It is easy to infer interactions, has strong scalability, can predict the behavior of any agent, and is consistent with BEV perception. However, the dense representation is limited by the fixed receptive field and resolution. This greatly affects the robust safety of prediction and makes it difficult to handle occlusions that may interact with driving strategies. Contrary to the dense per-pixel behavior probability modeling, the sparse representation paradigm predicts clusters of multi-agent anchored trajectories or intention distributions. It provides a multi-modal representation for each agent, enabling flexible modeling of various behavior uncertainties under spatial semantics. However, non-aligned behavior conflicts and mode collapse hinder its generation of compliant multi-agent modeling. Meanwhile, as the number of agents increases, the computational cost grows exponentially. When behavior modeling involves joint prediction planning (IPP), these problems particularly lead to unstable and slow behavior paradigm learning. Typical solutions through conditional prediction or game-theoretic reasoning usually result in non-strategic planning because these methods are difficult to ensure the compliance and stability of mutual behavior paradigms in multi-round (rollout) interactions. This requires re-modeling the future behavior patterns of agents to be compliant and meet the optimization goals of IPP.

[0087] In the natural decision-making process of humans, human drivers mainly determine future surrounding interaction behaviors rather than the specific states of surrounding agents for decision-making. Therefore, an effective strategy includes evaluating the impact of the behavior of surrounding agents on the planning strategy and reasoning about compliant future interactions. This introduces the core concept of the embodiments of the present invention, that is, multi-agent behaviors that conform to social norms and are consistent exhibit topological structures, which can be identified by extracting consistent or consensus interactions from future behaviors. Previous studies have used graph neural networks (GNNs) or Transformers to address this challenge through structural design or implicit relationship learning. Other studies have quantified the uncertainty of driving scenarios through topological properties. However, there is currently no method to explicitly formulate a coordinated and consistent supervised representation for the future behavior patterns of agents.

[0088] This embodiment presents a multi-agent future behavior representation called Behavior Topology (BeTop). The core of BeTop is an explicit representation of a topological supervision where multi-agent future interactions have consensus, and the establishment of an inference task to guide autonomous driving prediction and planning. BeTop is derived from braid theory, which infers consistent interaction paradigms in multiple trajectories under the transformation of knot space from intertwining. This enables BeTop to intuitively extract the forward intertwining (occupation) as a joint topology from the multi-agent future trajectories abstracted as knot trajectories, combining dense and sparse representations. With the help of BeTop, this embodiment introduces a Transformer-based collaborative learning model framework BeTopNet for joint prediction and planning (IPP) objectives. Among them, the synergistic decoder synergistically infers the behavior topology through an iterative decoding module strategy and generates a cluster of future trajectories. At the same time, the topology-guided local attention module embedded in each decoder layer adaptively utilizes the inferred BeTop topology prior to query the agent behavior semantics of future potential interactions for information aggregation. To alleviate the uncertainty of multi-agents in the driving scenario through the topology prior of BeTop, this embodiment designs and deploys a contingency planning paradigm to optimize the learning of the model. This embodiment designs an imitative contingency learning paradigm, which standardizes short-term planning with safety guarantees and solves the uncertainty of long-term planning through the joint prediction inferred by BeTop. Experimental results show that in large-scale public data scenarios, the consistency and accuracy of prediction and planning are enhanced, and state-of-the-art performance is achieved. The simulation test conducted in the proposed interaction scenario benchmark further highlights the planning ability of BeTopNet.

[0089] The technical route of the embodiment of the present invention mainly consists of three parts: First, establish a Behavior Topology (BeTop) representation based on braid theory and construct the corresponding topological inference task; Second, based on the topological inference task, establish a collaborative learning model BeTopNet and specific module implementation; Finally, introduce an imitative contingency learning paradigm to optimize the learning of the inference task.

[0090] First, construction of behavior topology representation:

[0091] Problem construction: Consider a driving scenario composed of N a multi-agents at the current time t = 0, and each driving scenario is represented as Consider the scenario graph M at the same time. The corresponding agent within the historical scope T h The states are respectively represented as the X of the autonomous vehicle (AV) 1 and those of other scenario agents Among them, for the future scope T f The plan Y 1 The goal is n ∈ [1, N a . The integrated prediction and planning aims to jointly predict the trajectories of agents and the ego-vehicle in the scenario

[0092] Topology construction: Using knot theory, which explores explicit representations conforming to multi-agent interactions from future trajectories Intuitively, it represents a transformation process According to each agent, local transformations are performed, and then the forward intertwined representations of each future trajectory are collected to represent future interactions. Formally, consider the knot group a Composed of N n Original knots σ Where each Represents a set of monotonically increasing functions Represents the Cartesian coordinate system Mapped to the future Lateral coordinate system of the agent Specifically, the function n In σ Is defined as 1 ≤ i, n ≤ N a , where b n And R n Represent the left-handed coordinate system transformation matrix of the local coordinates of agent A n In this way, the multi-agent joint interaction behavior is recognized as the knot intertwining of one trajectory with another Different from the implicit method relying on the future trajectory distance heuristic, each intertwining in the knot can be represented as an explicit interaction behavior response, distinguishing active interactions ( Causing other agents to yield and give way) and passive interactions ( Yielding to other agents) of future behaviors. To avoid difficult dynamic knot set reasoning, the multi-agent intertwining is reconstructed from the perspective of topological reasoning

[0093] The goal of Behavior Topology (BeTop) is to infer the topological graph of multi-agent future behaviors Specifically, the node topology Is constructed as the future trajectories of multi-agents. At the same time, the knot intertwining set Is reformulated as the edge topology of future interaction behaviors 1 ≤ i, j ≤ N a For each topological edge element e ij can be defined by two knot functions to evaluate the behavioral intertwining relationship of future Y i , Y j : Here, I(.) represents the function for judging the intersection of line segments under the horizontal coordinate. The inference task of the behavioral topology is further defined as:

[0094]

[0095] where represents the topological graph, represents the prediction node item, that is, the prediction knot, which is the position information of each agent sample, represents the set of prediction edges, and the set includes several topological edges e ij .

[0096] Node item The future states of multiple agents in are defined and optimized by Gaussian mixture (GMM). Edge topology is constructed as a probabilistic inference problem:

[0097]

[0098] where e ij represents the edge topology.

[0099] Next, a collaborative topological inference learning model is established:

[0100] This embodiment introduces a collaborative learning framework BeTopNet. Its main body is an encoder-decoder deep network based on Transformer as the backbone. By encoding the scene semantics X; M, the proposed network can use the designed collaborative decoder module to achieve simultaneous inference and guidance of BeTop. Edge topology The inference output head of and the joint prediction planning (IPP) output head of

[0101] Driver scene encoder: This embodiment follows the principle criterion centered on planning and constructs the model input using a scene-centered coordinate system. The scene feature attributes include the historical states X of multiple agents and the map vector input M, where this embodiment segments the length L m from the complete scene map m map segments of length N A , S M, and concatenated into scene features. S = [S A ; S M The Transformer encoder stack with local attention mechanism is directly used to capture the information interaction of historical local regions from the encoded scene semantics S A , S M .

[0102] Collaborative decoder: Based on the encoded scene features S A , S M , this embodiment focuses on the decoding strategy. This strategy requires: 1) reasoning about the co-edge topology of BeTop nodes interactively and simultaneously; 2) adaptively aggregating the future interaction agent features using the inferred edge topology prior. To this end, this embodiment constructs an iterative decoding strategy consisting of N Transformer decoder layers for all agents. To eliminate the uncertainty of the scene, a set of multimodal M decoding queries is initialized for the multi-agent future trajectories Meanwhile, the relative relationship feature S R is deployed as a topological feature through a multi-layer perceptron (MLP) for edge topology reasoning. Next, this embodiment designs a dual information flow for the iterative decoding process of future trajectories and future topologies . The decoding process of agent A n at layer l is as follows:

[0103]

[0104]

[0105] Among them, the future trajectories and the interactive topology e ij ∈ε in BeTop are decoded in a collaborative manner. The inferred edge topology is obtained by the topology decoder through the query ; this embodiment further uses as a prior to establish a Transformer decoder with a topology-guided local attention mechanism and the inferred output nodes.

[0106] Topology-guided local attention: Querying the semantic features of all agents in the full scene will lead to inconsistencies in future interactions and sparse attention distributions. This prompts this embodiment to design local attention to infer the edge topology as a prior. Specifically, this embodiment retrieves the top K indices prior to the inferred edge topology for the final behavioral interaction with A n . The indices with interactions Directly used for adaptively aggregating S A For local cross-attention modeling. This process is formed as follows:

[0107]

[0108] Wherein, Subsequent decoder layers have interactive agent feature aggregations represented as

[0109] Inference output head: Given the decoded features of each layer In this embodiment, additional inference output heads are added, and these inference heads correspond to the edges of the behavior topology graph and the node topology output. Among them, the topology output head, the planning output head, and the prediction output head (IPP) are simultaneously output by the stacked MLP when inferring the BeTop result. For each layer of agent A n , the inference head decodes the future trajectory state To generate components modeled by a Gaussian mixture distribution (GMM) (including μ x , μ y , σ x , σ y , ρ), where the mixture distribution fraction And generate the edge topology

[0110] Finally, imitate contingency planning learning:

[0111] BeTopNet learns to imitate the end-to-end goal of human multi-agent behavior, integrating scenario uncertainty compliance behavior through contingency planning.

[0112] The imitation goal is first established by regulating the multi-agent behavior state While maximizing their interaction distribution The imitation goal is the negative log-likelihood (NLL) of the best inference component m* closest to the true value; the behavior distribution of the edge topology is calculated by binary cross-entropy (BCE). To integrate Compliant behavior learning in multi-agent scenario uncertainty, contingency planning has proven to be a suitable solution. By jointly predicting the immediate safety plan τ M Connected to the branch planning trajectory cluster It can defer uncertain decisions and ensure the safety of actual planning. Making direct joint predictions may lose diversity, while the inferred behavior topology can serve as a suitable medium to refine future interactive agents and achieve effective joint combinations. Given the imitation ego-vehicle planning output And the planning branch time is t b∈(1, T f ), the learning of joint emergency planning first utilizes all marginal predictions to guide safe short-term planning and a set of M branch plans guided by joint prediction , which is defined as: where maxC

[0113]

[0114] represents the worst-case cost function (joint prediction M and scenario probability are recombined from K subsets of interactive multi-agent marginal predictions for fast indexing from the sorted AV topology M , and it is described by the joint cost function C that guides the branch planning strategy. Specifically, both types of cost functions are defined by repulsive potential fields to reduce the proximity of the planning to their respective prediction paradigms. J The technical solution implemented in the present invention introduces the concept of Behavior Topology (BeTop), which is a multi-agent future behavior paradigm for topological reasoning that explicitly supervises the consistent future collaborative interaction behavior of the Joint Prediction Planning (IPP) system. In addition, a collaborative learning framework BeTopNet is designed to provide joint planning and prediction guided by topological reasoning. Topology-guided local attention and simulated emergency planning can solve scenario compliance and multi-agent uncertainty, and this solution has achieved the most excellent performance in both planning strategies and prediction accuracy.

[0115] Currently, there is no method to explicitly formulate a collaborative supervisory representation for the future behavior patterns of agents. In this embodiment, a topological form of supervisory representation with consensus for multi-agent future interactions is explicitly established, namely the proposed Behavior Topology. The technical solution of this embodiment realizes the collaborative reasoning of the behavior topology graph by designing the model BeTopNet, and at the same time uses the interactive edge node reasoning prior to enhance the state generation of prediction planning, and uses the imitation emergency planning learning paradigm to organically integrate the topological reasoning results to solve uncertainty.

[0116]

[0117] Embodiment 2

[0118] Figure 2 ​It is a schematic structural diagram of a future behavior topology inference device for autonomous driving multi - agents in an embodiment of the present invention. This embodiment is applicable to the situation of future behavior topology inference for autonomous driving multi - agents. The device can be implemented in software and / or hardware, and can be integrated into any device that provides the function of future behavior topology inference for autonomous driving multi - agents, such as Figure 2 As shown, the future behavior topology inference device for autonomous driving multi - agents specifically includes: an acquisition module 201 and an input module 202.

[0119] Among them, the acquisition module 201 is used to acquire the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located;

[0120] The input module 202 is used to input the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology inference result. Among them, the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology inference result sample corresponding to the agent set sample.

[0121] Optionally, the input module 202 includes:

[0122] A building unit, used to build an initial model;

[0123] A first input unit, used to input the agent set sample in the training sample set into the initial model to obtain a predicted future behavior topology inference result corresponding to the agent set sample;

[0124] A first determination unit, used to determine a fitness value based on a preset fitness function according to the predicted future behavior topology inference result corresponding to the agent set sample;

[0125] A training unit, used to train the parameters of the initial model according to a loss function formed by the predicted future behavior topology inference result corresponding to the agent set sample and the future behavior topology inference result sample corresponding to the agent set sample if the fitness value does not meet the preset target;

[0126] An execution unit, used to return and execute the operation of inputting the agent set sample in the training sample set into the initial model to obtain a predicted future behavior topology inference result corresponding to the agent set sample until the fitness value meets the preset target, or the number of iterations is greater than or equal to a preset number of times, to obtain a collaborative learning model.

[0127] Optionally, the input module 202 includes:

[0128] A second input unit, configured to input the agent set samples in the training sample set into the initial model;

[0129] A first acquisition unit, configured to acquire the position information of each agent sample in the agent set sample;

[0130] A construction unit, configured to construct a topological graph according to the position information of each agent sample;

[0131] A third input unit, configured to input the topological graph into the encoder - decoder in the initial model to obtain the predicted future behavior topological inference result corresponding to the agent set sample.

[0132] Optionally, the construction unit is specifically configured to:

[0133] Based on the topological knot theory, using each agent sample as a knot, generate a knot interweaving set according to the position information of each agent sample;

[0134] Determine a topological edge set according to the knot interweaving set, where each topological edge represents the future interaction behavior between the two agents corresponding to the topological edge;

[0135] Construct a topological graph according to each knot and each topological edge.

[0136] Optionally, the apparatus further includes:

[0137] A second acquisition unit, configured to acquire the knot function corresponding to each knot;

[0138] A second determination unit, configured to determine the edge topological function of the topological edge corresponding to each two knots according to every two knot functions;

[0139] A third determination unit, configured to determine the preset fitness function according to the knot function corresponding to each knot and the edge topological function of each topological edge.

[0140] Optionally, the third input unit is specifically configured to:

[0141] Input the topological graph into the encoder in the initial model to obtain encoded scene features;

[0142] Input the encoded scene features into the decoder in the initial model to obtain the predicted future behavior topological inference result corresponding to the agent set sample.

[0143] The above - mentioned product can execute the autonomous driving multi - agent future behavior topological inference method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the method.

[0144] Embodiment III

[0145] Figure 3 FIG. 1 shows a schematic structural diagram of an electronic device 30 that can be used to implement an embodiment of the present invention. The electronic device is intended to represent various forms of digital computers, such as, laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0146] As Figure 3 shown, the electronic device 30 includes at least one processor 31, and a memory communicatively connected to the at least one processor 31, such as a read-only memory (ROM) 32, a random access memory (RAM) 33, etc. The memory stores a computer program executable by the at least one processor. The processor 31 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 32 or the computer program loaded from the storage unit 38 into the random access memory (RAM) 33. In the RAM 33, various programs and data required for the operation of the electronic device 30 can also be stored. The processor 31, the ROM 32, and the RAM 33 are connected to each other through a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.

[0147] A plurality of components in the electronic device 30 are connected to the I / O interface 35, including: an input unit 36, such as a keyboard, a mouse, etc.; an output unit 37, such as various types of displays, speakers, etc.; a storage unit 38, such as a magnetic disk, an optical disk, etc.; and a communication unit 39, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 39 allows the electronic device 30 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0148] The processor 31 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 31 include but are not limited to a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 31 executes the various methods and processes described above, such as the autonomous driving multi-agent future behavior topology inference method:

[0149] Obtain the position information of the current vehicle and the position information of each agent in the set of agents included in the scenario where the current vehicle is located;

[0150] Input the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology inference result, where the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology inference result sample corresponding to the agent set sample.

[0151] In some embodiments, the multi-agent future behavior topology inference method for autonomous driving can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as storage unit 38. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 30 via ROM 32 and / or communication unit 39. When the computer program is loaded into RAM 33 and executed by the processor 31, one or more steps of the multi-agent future behavior topology inference method for autonomous driving described above can be executed. Alternatively, in other embodiments, the processor 31 can be configured to execute the multi-agent future behavior topology inference method for autonomous driving in any other suitable manner (e.g., by means of firmware).

[0152] The various embodiments of the systems and techniques described above in this document 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-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: implemented in one or more computer programs, which can be executed and / or interpreted on a programmable system including at least one programmable processor, and the programmable processor can be a dedicated or general-purpose programmable processor, and can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0153] The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to the processor of a general-purpose computer, a dedicated computer, or other programmable data processing devices, such that when the computer program is executed by the processor, the functions / operations specified in the flowchart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0154] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0155] To provide for interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the electronic device. Other kinds of devices can also be used to provide for 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 acoustic input, speech input, or tactile input).

[0156] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0157] A computing system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs that run on respective computers and have a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0158] In one embodiment, the embodiment of the present invention further includes a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the future behavior topology inference method for autonomous driving multi-agent in any embodiment of the present invention.

[0159] In the process of implementation, the computer program product can write computer program code for performing the operations of the present invention in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0160] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is made herein.

[0161] The above specific embodiments do not constitute a limitation to the protection scope of the present invention. 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 principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for topological reasoning of future behaviors of multi-agents in autonomous driving, characterized in that: include: Acquire the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located; Inputting the position information of the current vehicle and the position information of each agent into the collaborative learning model to obtain the future behavior topology reasoning result, wherein the collaborative learning model is obtained by iteratively training the initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology reasoning result sample corresponding to the agent set sample; The initial model is iteratively trained through the training sample set, including: Build an initial model; Inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topological reasoning results corresponding to the agent set samples; Based on a preset fitness function, determining a fitness value according to a predicted future behavior topological reasoning result corresponding to the agent set sample; If the fitness value does not meet the preset target, the parameters of the initial model are trained according to the loss function formed by the predicted future behavior topology reasoning result corresponding to the agent set sample and the future behavior topology reasoning result sample corresponding to the agent set sample; Return to executing the operation of inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topology reasoning results corresponding to the agent set samples, until the fitness value meets the preset target, or the number of iterations is greater than or equal to the preset number, to obtain a collaborative learning model; The step of inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topology reasoning results corresponding to the agent set samples includes: Inputting the agent set samples in the training sample set into the initial model; Obtaining location information of each agent sample in the agent set sample; Construct a topological map based on the location information of each agent sample; Inputting the topological graph into the codec in the initial model to obtain the predicted future behavior topological reasoning result corresponding to the agent set sample; Among them, a topological map is constructed according to the location information of each agent sample, including: Based on the topological knot theory, each agent sample is regarded as a knot, and a knot interweaving set is generated according to the position information of each agent sample; Determine a topological edge set according to the knot interweaving set, wherein each topological edge represents a future interaction behavior between two agents corresponding to the topological edge; A topological graph is constructed based on each knot and each topological edge.

2. The method according to claim 1, characterized in that Also includes: Get the knot function corresponding to each knot; Determine the edge topology function of the topological edge corresponding to each two knots according to each two knot functions; The preset fitness function is determined according to the knot function corresponding to each knot and the edge topology function of each topological edge.

3. The method according to claim 1, characterized in that Inputting the topological graph into the codec in the initial model to obtain the predicted future behavior topological reasoning result corresponding to the agent set sample, including: Inputting the topological map into an encoder in the initial model to obtain encoded scene features; The encoded scene features are input into the decoder in the initial model to obtain the predicted future behavior topology reasoning results corresponding to the agent set samples.

4. A device for topological reasoning of future behaviors of multi-agents in autonomous driving, characterized in that: include: An acquisition module, used to acquire the position information of the current vehicle and the position information of each agent in the set of agents included in the scene where the current vehicle is located; An input module is used to input the position information of the current vehicle and the position information of each agent into a collaborative learning model to obtain a future behavior topology reasoning result, wherein the collaborative learning model is obtained by iteratively training an initial model with a training sample set, and the training sample set includes: an agent set sample and a future behavior topology reasoning result sample corresponding to the agent set sample; Wherein, the input module includes: Establishing unit, used to establish initial model; A first input unit, used for inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topological reasoning results corresponding to the agent set samples; A first determination unit, configured to determine a fitness value according to a predicted future behavior topology reasoning result corresponding to the agent set sample based on a preset fitness function; A training unit, configured to train the parameters of the initial model according to a loss function formed by the predicted future behavior topology reasoning result corresponding to the agent set sample and the future behavior topology reasoning result sample corresponding to the agent set sample if the fitness value does not meet the preset target; An execution unit, used to return to execute the operation of inputting the agent set samples in the training sample set into the initial model to obtain the predicted future behavior topological reasoning results corresponding to the agent set samples, until the fitness value meets the preset target, or the number of iterations is greater than or equal to the preset number, to obtain a collaborative learning model; Wherein, the input module includes: A second input unit, used to input the agent set samples in the training sample set into the initial model; A first acquisition unit, configured to acquire location information of each agent sample in the agent set sample; A construction unit, used for constructing a topological map according to the location information of each agent sample; A third input unit is used to input the topological map into the codec in the initial model to obtain the predicted future behavior topological reasoning result corresponding to the agent set sample; Wherein, the construction unit is specifically used for: Based on the topological knot theory, each agent sample is regarded as a knot, and a knot interweaving set is generated according to the position information of each agent sample; Determine a topological edge set according to the knot interweaving set, wherein each topological edge represents a future interaction behavior between two agents corresponding to the topological edge; A topological graph is constructed based on each knot and each topological edge.

5. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the autonomous driving multi-agent future behavior topology reasoning method described in any one of claims 1-3.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the autonomous driving multi-agent future behavior topology reasoning method according to any one of claims 1 to 3 when executed.

7. A computer program product, comprising a computer program, which, when executed by a processor, implements the autonomous driving multi-agent future behavior topology reasoning method according to any one of claims 1-3.

Images