Autonomous driving test method, equipment and medium based on multi-agent cluster confrontation

Abstract

The present invention relates to an automatic driving test method, equipment and medium based on multi-agent cluster confrontation, the method comprising: S1, deduce the conflict topological relationship diagram between the tested automatic driving vehicle and the intelligent agent; S2, deduce the feasible planning space of the tested automatic driving vehicle according to the conflict topological relationship diagram; S3, according to the correlation between the individual interests of the intelligent agent and the test effect of the multi-agent system cluster confrontation, establish a multi-agent cluster confrontation model based on potential game in the feasible planning space, and solve the optimal confrontation strategy of the multi-agent system against the tested automatic driving vehicle; wherein, the confrontation intensity is introduced into the multi-agent cluster confrontation model, and the confrontation intensity is adaptively adjusted according to the actual response of the tested automatic driving vehicle; S4, repeat S1 to S3 until the confrontation task is completed. Compared with the prior art, the present invention realizes cluster confrontation with dynamically adjustable confrontation intensity, which can fully test the decision-making ability of the automatic driving vehicle in complex interactive scenarios.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving testing technology, and in particular to an autonomous driving testing method, equipment and medium based on multi-agent cluster confrontation. Background Art

[0002] As autonomous driving technology matures, intelligent connected vehicles have moved from the first half of the competition, which focused on early technology, to the second half, which is dominated by commercialization. However, in complex interactive environments, autonomous vehicles still behave overly conservatively, indicating that the current capabilities of autonomous vehicles in high-density environments need to be further improved and tested.

[0003] Current testing of autonomous vehicles often involves a small number of environmental vehicles, and generally only tests the capabilities of autonomous vehicles by injecting faults into a single environmental vehicle. This method not only produces dangerous scenarios that are almost non-existent in the real world, but also fails to take into account that driving tasks on open roads are often affected by the coupling of multiple objects, resulting in inaccurate and unreliable test results for autonomous vehicles, which in turn causes current autonomous vehicles to "not adapt well" to open roads. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide an autonomous driving test method, equipment and medium based on multi-agent cluster confrontation, to achieve cluster confrontation with dynamically adjustable confrontation intensity, and to fully test the decision-making ability of autonomous driving vehicles in complex interactive scenarios.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] According to a first aspect of the present invention, a method for testing autonomous driving based on multi-agent cluster confrontation is provided, the method comprising the following steps:

[0007] Step S1, deriving a conflict topology relationship diagram between the tested autonomous driving vehicle and the intelligent agent according to the test scene road topology and the test object conflict relationship;

[0008] Step S2: deriving a feasible planning space of the autonomous driving vehicle under test according to the conflict topology relationship diagram;

[0009] Step S3: According to the correlation between the individual interests of the intelligent agents and the cluster confrontation test effect of the multi-agent system, a multi-agent cluster confrontation model based on potential game is established in the feasible planning space to solve the optimal confrontation strategy of the multi-agent system against the tested autonomous driving vehicle; wherein, in the multi-agent cluster confrontation model, the confrontation intensity is introduced to characterize the relative weight of the individual benefits of the intelligent agents and the cluster confrontation test effect of the multi-agent system, and the confrontation intensity is adaptively adjusted according to the actual response of the tested autonomous driving vehicle;

[0010] Step S4: Repeat steps S1 to S3 until the confrontation task is completed.

[0011] Preferably, the step S1 specifically includes:

[0012] Based on the vehicle status information and test map information of the multi-agent vehicle group {Veh1, Veh2...Vehn} and the tested autonomous vehicle Veh0, deduce whether there is a spatial conflict between the tested autonomous vehicle and the multi-agents in the environment, and between the multi-agents;

[0013] Graph theory is used to describe the topological relationship of vehicle conflicts, and the conflict topological relationship diagram G = (V, E) between the tested autonomous vehicle and the intelligent agent is obtained; where V represents the set of vehicles, and the position of vehicle Vehn at time t is p b =(x n (t),y n (t)), with a velocity of v n (t); E represents the set of edges. For edge e ij , according to the current position p of the vehicle Vehi i and the current position p of vehicle Vehj j Make inferences. If there is a conflict in space, record e ij =1, otherwise it is recorded as 0.

[0014] Preferably, step S2 specifically includes: reducing the two-dimensional conflict problem to a one-dimensional following problem according to the conflict topology relationship diagram through a virtual lane projection method, and deducing the feasible domain of the tested autonomous driving vehicle under the confrontation of the multi-agent system by establishing the relationship between the longitudinal motion of the vehicle and time, thereby characterizing the feasible planning space of the tested autonomous driving vehicle under the current conflict topology relationship.

[0015] Preferably, in step S3, according to the correlation between the individual interests of the agents and the test results of the multi-agent system cluster confrontation, a multi-agent cluster confrontation model based on potential game is established in the feasible planning space, and the expression is:

[0016]

[0017] Where: Indicates that agent i is in the adversarial strategy The cluster confrontation effect of the multi-agent system under the condition, U is the feasible strategy space, V is the set of multi-agent vehicles; represents the cluster confrontation effect of multi-agent system under any initial confrontation strategy of agent i; Indicates that agent i is in the adversarial strategy The individual benefits under represents the individual benefit of agent i under the initial adversarial strategy; a i is the acceleration of agent i, p i is the position of agent i, and p0 is the position of the autonomous driving vehicle under test.

[0018] Preferably, the function expression of the individual benefit of the intelligent agent in step S3 is:

[0019]

[0020] Where: represents the driving reward of agent i at time t, a i is the acceleration of agent i, p i is the position of agent i, is the distance between agent i and the end point, is the acceleration of agent i; represents the adversarial reward of agent i at time t, It represents the time difference between the intelligent agent i and the tested autonomous driving vehicle Veh0 when they reach the conflict point at time t, represents the distance between agent i and the conflict point, represents the speed of agent i, Indicates the distance between the tested autonomous vehicle and the conflict point, represents the speed of the tested autonomous vehicle, p0 is the position of the tested autonomous vehicle; θ is the strength of the confrontation, which is characterized by the relative weight of the individual benefit of the agent and the effect of the multi-agent system cluster confrontation test;

[0021] The multi-agent system cluster confrontation test effect, the function expression is:

[0022]

[0023] Where: γ represents the benefit reduction coefficient; T is the planning step length; is the contribution of agent i in the confrontation.

[0024] Preferably, the contribution of the agent i in the confrontation The calculation expression is:

[0025]

[0026] Where c represents the possible combination of agent i in the multi-agent system, and P(c)-P(c-{i}) represents the marginal utility generated by agent i in combination c.

[0027] Preferably, the adversarial strength θ uses a reverse update algorithm to adaptively and dynamically update the relative weight, and the update expression is:

[0028]

[0029] Where: θ t-1 is the strength of the confrontation at time t-1, μ represents the update rate of the strength of the confrontation, Estimated effect of the tested autonomous driving vehicle facing cluster confrontation.

[0030] Preferably, in step S3, a minimum sequence quadratic programming method is used to solve the optimal Nash equilibrium confrontation strategy of the multi-agent system against the tested autonomous driving vehicle.

[0031] According to a second aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, any one of the methods described above is implemented.

[0032] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, any one of the methods described above is implemented.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] 1) Compared with the traditional autonomous driving test in which a single environmental vehicle is controlled to suddenly fail and then a specific single function of the tested autonomous driving vehicle is detected, the present invention is applicable to all scenarios with driving conflicts through the deduction of the conflict topology relationship diagram and the feasible planning space. It has good scalability, can meet the testing requirements of massive scenarios, and test the overall capabilities of the autonomous driving decision-making module.

[0035] 2) According to the correlation between the individual interests of intelligent agents and the test results of multi-agent system cluster confrontation, the present invention establishes a multi-agent cluster confrontation model based on potential game in the feasible planning space, which more realistically reflects the driving conflict relationship of autonomous driving vehicles and facilitates the accurate solution of the optimal confrontation strategy of the multi-agent system against the tested autonomous driving vehicle.

[0036] 3) The present invention can dynamically and adaptively adjust the confrontation strength of the multi-agent system according to the actual behavior of the tested autonomous driving vehicle and when the decision-making algorithm of the tested autonomous driving vehicle is opaque, so as to realize tests of different confrontation strengths and further fully test the decision-making ability of the tested autonomous driving vehicle in different situations.

[0037] 4) The present invention obtains the contribution of each intelligent agent in the confrontation by calculating the marginal utility of the vehicle. By enhancing the confrontation strength of the key intelligent agents, the feasible planning space of the tested autonomous driving vehicle can be compressed in a more targeted manner, further improving the reliability and accuracy of solving the optimal confrontation strategy.

[0038] 5) Using the weighted reverse update algorithm, the adversarial strength is adaptively updated according to the real-time behavior of the autonomous vehicle under test and the intelligent system, which can achieve reliable testing of different adversarial strengths.

[0039] 6) The present invention adopts the minimum sequence quadratic programming method to solve the multi-agent cluster confrontation model based on potential game, and obtains the optimal confrontation strategy of the multi-agent system against the tested autonomous driving vehicle, meeting the requirements of low computing resource consumption and high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is the overall flow chart of the present invention;

[0041] Figure 2 A schematic diagram of the initial state of a vehicle and a test environment in the example of the present invention;

[0042] Figure 3 A schematic diagram of a conflicting topological relationship in an example of the present invention;

[0043] Figure 4 is a schematic diagram of a virtual lane projection in the example of the present invention;

[0044] Figure 5 It is a schematic diagram of the spatiotemporal feasible domain of the autonomous driving under test in the example described in the present invention; DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0046] Example

[0047] This embodiment provides an autonomous driving test method based on multi-agent cluster confrontation, the method comprising the following steps:

[0048] Initialize the test scenario. At the beginning of the test, initialize the vehicle status information of the multi-agent vehicle group {Veh1, Veh2...Veh4} and the tested autonomous vehicle Veh0, such as Figure 2As shown, the multi-agent vehicle group and the autonomous driving vehicle under test are started.

[0049] Step 1: Create a conflict topology diagram.

[0050] Step 11: Based on the vehicle status information and test map information of the multi-agent vehicle group {Veh1, Veh2...Vehn} and the autonomous driving vehicle under test Veh0, deduce whether there is a spatial conflict between the autonomous driving vehicle under test and the multi-agents in the environment, and between the multi-agents.

[0051] Step 12: Describe the topological relationship of vehicle conflicts based on graph theory. V represents the set of vehicles. For vehicle Vehn, its position at time t is p n =(x n (t),y n (t)), with a velocity of v n (t), E represents the set of edges, which is an n×n adjacency matrix. For vehicles Vehi and Vehj, according to their current positions p i and p j Make inferences, if there is a conflict, record ij =1, otherwise it is recorded as 0, and then the conflict relationship of the vehicles can be obtained. Finally, the conflict relationship of the vehicles can be described by the graph G = (V, E), such as Figure 3 shown.

[0052] Step 2: Modeling of the spatiotemporal feasible planning space of the autonomous driving under test.

[0053] Step 21: Considering that the conflict between multiple vehicles often requires the simultaneous consideration of the lateral and longitudinal trajectory planning of the vehicle, this problem is a non-convex problem with high computational complexity and large amount of calculation. Therefore, the virtual lane projection method is introduced here to reduce the dimension of the two-dimensional vehicle conflict to the longitudinal planning of the one-dimensional scene according to the conflict topological relationship of the vehicle, which greatly simplifies the amount of calculation and ensures the real-time confrontation needs, such as Figure 4 As shown, since Veh4 is not connected to the tested vehicle Veh0 in the conflict topology map, no virtual projection is performed. According to the sequence of conflict points on the planned trajectory of Veh1, Veh2, Veh3 and the tested vehicle Veh0, they are projected onto the longitudinal planning line of Veh0 in order.

[0054] Step 22: Based on the position and speed information of the autonomous driving vehicle under test, the correlation between the longitudinal motion and time of the autonomous driving vehicle under test is established, and the feasible planning space of the autonomous driving vehicle under test is inferred by combining the results of the virtual lane projection and the vehicles in the multi-agent system and the vehicles in the multi-agent system, such as Figure 5As shown, in this example, the autonomous driving vehicle Veh0 under test conflicts with Veh1, Veh2, and Veh3 successively, and its feasible planning space must avoid the space already occupied by the above-mentioned intelligent agents.

[0055] Step 3: Multi-agent cluster adversarial modeling.

[0056] Step 31: The individual benefit of the agent is established, where the individual benefit of the agent mainly includes two parts, the specific forms are as follows:

[0057]

[0058] Among them, R vehi Represents the individual benefit of agent i, which mainly includes the driving reward at time t and confrontation rewards Driving Rewards It mainly includes two aspects: efficiency and comfort. i and is the agent’s acceleration and distance to the endpoint, characterizing the efficiency of the driving process; is the acceleration of the agent, which represents the comfort of the driving process. It means that the agent can get rewards from the confrontation with the tested autonomous vehicle, specifically the time difference between the agent and the tested autonomous vehicle Veh0 when they reach the conflict point The smaller the value, the more severe the conflict. At this time, the agent can gain benefits from the confrontation. represents the distance between the agent and the conflict point, Indicates the current speed of the agent.

[0059] Step 32: Establishment of the effect function of the multi-agent system cluster confrontation test. After obtaining the individual benefits of the agent, the joint benefit equation of the individual benefits of the agent and the multi-agent system cluster confrontation test effect P is established through the potential game model. First, the definition of the potential game is as follows:

[0060]

[0061] in Indicates that the confrontation strategy of agent i is When , U is the feasible strategy space. When the individual benefits of the agents and the confrontation effect of the multi-agent system satisfy the above inequality, it is considered that the confrontation between the multi-agent system and the tested autonomous driving vehicle can be called a potential game problem.

[0062] Furthermore, the effect function of the multi-agent system cluster confrontation test is as follows:

[0063]

[0064] Among them, γ t It represents the discount coefficient of the benefit, which characterizes the preference relationship between the agent's recent benefit and future benefit.

[0065] Step 33: Solve the optimal confrontation strategy for Nash equilibrium. After obtaining the objective function of the confrontation effect of the multi-agent system, considering that there is an absolute value term in this problem, it is a non-convex optimization and can be solved by the minimum sequence quadratic programming method. The main steps include: converting the objective function and constraints into a continuously differentiable form, calculating the objective function and its gradient, as well as the constraints and their Jacobian matrix; in each iteration, solving a convex optimization subproblem to update the current solution based on the current solution and the gradient information of the objective function; in each iteration, check whether the change in the objective function value is less than the pre-set convergence tolerance. If the change is less than the tolerance, it is considered that the algorithm has converged, and finally the optimal confrontation strategy for the Nash equilibrium of the multi-agent system against the autonomous driving vehicle under test is obtained.

[0066] Step 4: Dynamically adjust the strength of the confrontation.

[0067] Step 41: Quantitative representation of the strength of confrontation. In the establishment of the individual benefits of the agent in step 31, the benefits mainly include driving rewards and confrontation rewards. The relative weight θ is introduced to represent the agent's preference for driving rewards and confrontation rewards. When the corresponding weight of the confrontation reward is high, the main goal of the agent changes from leaving the test scene quickly and safely to generating stronger confrontation with the tested autonomous driving. Therefore, the formula of the individual benefits of the agent and the confrontation effect function of the multi-agent system is updated as follows:

[0068]

[0069] By adjusting the relative weight θ, tests with different adversarial intensities can be implemented, thereby verifying the decision-making ability of the autonomous driving under test in different situations.

[0070] Step 42: Identification of key agents. There are some key agents in the multi-agent system that are in strong conflict with the autonomous vehicle under test. By calculating the marginal utility of the vehicle, we can get the contribution of each agent in the cluster system in the confrontation. By enhancing the confrontation strength of the key agents, we can more specifically compress the feasible planning space of the autonomous vehicle under test in step 2. Agent i contributes in the confrontation The calculation of is as follows:

[0071]

[0072] Among them, c represents the possible combination of agent i in the multi-agent system, and P(c)-P(c-{i}) represents the marginal utility generated by agent i in combination c. After obtaining the confrontation contribution, the agent with higher contribution is regarded as the key agent confronting the tested autonomous vehicle at time t, and the possible confrontation effect is considered during optimization. Therefore, the formula of the confrontation effect function of the multi-agent system is updated as follows:

[0073]

[0074] Step 43: Adaptive dynamic adjustment of the strength of the confrontation.

[0075] Since the decision-making algorithm of the tested autonomous vehicle is unknown to the multi-agent cluster confrontation system, the cluster confrontation should be loaded from nothing to something, from weak to strong, in order to test the boundaries of autonomous driving decision-making ability. Use the weight reverse update algorithm to update the confrontation intensity according to the real-time behavior of the tested autonomous vehicle and the real-time total objective function of the cluster system. The formula is as follows:

[0076]

[0077] Among them, μ represents the update rate of the confrontation strength. The larger the value, the faster the confrontation strength changes. It represents the estimated effect of the tested autonomous driving vehicle facing cluster confrontation. When this value is less than the actual confrontation effect P, it is considered that the tested autonomous driving vehicle has a low confrontation intensity θ t-1 The performance is good, so the strength of the next moment is increased. t , otherwise it will decrease, realizing adaptive dynamic adjustment of the confrontation intensity.

[0078] Step 5: Loop through the steps. Repeat steps 1 and 2 to correct the conflict topology and infer the space-time feasible domain, step 3 to search for the optimal confrontation strategy for Nash equilibrium, and step 4 to dynamically adjust the confrontation intensity until the confrontation task is completed.

[0079] After the test, the performance of the tested autonomous vehicle is comprehensively evaluated from multiple aspects such as safety and efficiency based on the trajectory data of the tested autonomous vehicle and the multi-agent cluster. For safety, the Post Encroachment Time (PET) is introduced. The Post Encroachment Time specifically refers to the time difference between the rear of the front vehicle and the front of the rear vehicle entering the conflict zone, where the conflict zone can be roughly regarded as the actual trajectory intersection of the tested autonomous vehicle and other vehicles. The specific form of the Post Encroachment Time is as follows:

[0080] PET=t front -t rear

[0081] Among them, t frontRefers to the time when the rear end of the front vehicle actually leaves the conflict area in the conflict relationship, t rear Refers to the time when the front of the rear vehicle actually enters the conflict zone in a conflict relationship. Generally speaking, PET < 0.7s is considered a serious conflict event, 0.7s ≤ PET < 1.3s is considered a general conflict event, 1.3s ≤ PET < 2.25s is considered a common conflict event, and PET ≥ 2.25s is considered a potential conflict event. The PET distribution of the tested autonomous vehicle under different confrontation intensities θ is collected, and the safety of the decision-making of the tested autonomous vehicle can be analyzed according to the composition ratio of different types of conflict events.

[0082] In terms of efficiency, travel delays are introduced and the difference between the actual passing time of the autonomous vehicle under test and the free flow environment is calculated. The specific form is as follows:

[0083] t delay =t actual -t free

[0084] Among them, t delay Refers to the delay time of the tested autonomous vehicle, t actual Refers to the time it takes for the tested autonomous vehicle to actually pass through the entire test area, t free Refers to the time required for the autonomous driving vehicle under test to pass through the test area in a free-flow environment.

[0085] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0086] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0087] The processing unit performs the various methods and processes described above, such as method steps 1 to 5. For example, in some embodiments, method steps 1 to 5 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more of the method steps 1 to 5 described above may be executed. Alternatively, in other embodiments, the CPU may be configured to execute method steps 1 to 5 in any other appropriate manner (e.g., by means of firmware).

[0088] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip systems (SOCs), load programmable logic devices (CPLDs), and the like.

[0089] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0090] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0091] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An autonomous driving test method based on multi-agent cluster confrontation, characterized in that: The method comprises the following steps: Step S1: Based on the test scene road topology and the test object conflict relationship, deduce the conflict topology relationship diagram between the tested autonomous driving vehicle and the intelligent agent, specifically including: Based on the vehicle status information and test map information of the multi-agent vehicle group {Veh1, Veh2...Vehn} and the tested autonomous vehicle Veh0, deduce whether there is a spatial conflict between the tested autonomous vehicle and the multi-agents in the environment, and between the multi-agents; Graph theory is used to describe the topological relationship of vehicle conflicts, and the conflict topological relationship diagram between the tested autonomous vehicle and the intelligent agent is obtained. ;in, Represents a collection of vehicles, and the vehicle Vehn is The time position is , the speed is ; represents the set of edges. , based on the current position of the vehicle Vehi and the current position of the vehicle Vehj Infer and if there is a conflict in space, record , otherwise it is recorded as 0; Step S2: deriving a feasible planning space of the autonomous driving vehicle under test according to the conflict topology relationship diagram; Step S3: According to the correlation between the individual interests of the intelligent agents and the cluster confrontation test effect of the multi-agent system, a multi-agent cluster confrontation model based on potential game is established in the feasible planning space, and the optimal confrontation strategy of the multi-agent system against the tested autonomous driving vehicle is solved; wherein, in the multi-agent cluster confrontation model, the confrontation intensity is introduced to characterize the relative weight of the individual benefits of the intelligent agents and the cluster confrontation test effect of the multi-agent system, and the confrontation intensity is adaptively adjusted according to the actual response of the tested autonomous driving vehicle; According to the correlation between the individual interests of the intelligent agent and the test results of the multi-agent system cluster confrontation, a multi-agent cluster confrontation model based on potential game is established in the feasible planning space, and the expression is: , Where: Indicates that the agent i is in the adversarial strategy The cluster confrontation effect of multi-agent system under is the feasible strategy space; represents the cluster confrontation effect of multi-agent system under any initial confrontation strategy of agent i; Indicates that the agent i is in the adversarial strategy The individual benefits of the intelligent agent under represents the individual benefit of agent i under the initial adversarial strategy; is the acceleration of agent i, is the position of agent i, is the position of the autonomous driving vehicle under test; The individual benefit of the intelligent agent is expressed as follows: ， , , Where: express The driving reward of agent i at time is the acceleration of agent i, is the position of agent i, is the distance between agent i and the end point, is the acceleration of agent i; express The adversarial reward of agent i at time express The time difference between the intelligent agent i and the tested autonomous driving vehicle Veh0 when they reach the conflict point, represents the distance between agent i and the conflict point, represents the speed of agent i, Indicates the distance between the tested autonomous vehicle and the conflict point, represents the speed of the autonomous driving vehicle under test, is the position of the autonomous driving vehicle under test; The relative weight of the individual benefit of the agent and the cluster confrontation test effect of the multi-agent system is represented for the confrontation intensity; The multi-agent system cluster confrontation test effect, the function expression is: , Where: represents the revenue reduction factor; is the planning step length; The contribution of agent i in the confrontation; Step S4: Repeat steps S1 to S3 until the confrontation task is completed.

2. The method for testing autonomous driving based on multi-agent cluster confrontation according to claim 1, characterized in that: The step S2 specifically includes: reducing the two-dimensional conflict problem to a one-dimensional following problem through the virtual lane projection method according to the conflict topology relationship diagram, and deducing the feasible domain of the tested autonomous driving vehicle under the confrontation of the multi-agent system by establishing the relationship between the longitudinal motion of the vehicle and time, thereby characterizing the feasible planning space of the tested autonomous driving vehicle under the current conflict topology relationship.

3. The method for testing autonomous driving based on multi-agent cluster confrontation according to claim 1, characterized in that: The contribution of the agent i in the confrontation , the calculation expression is: , Where: represents the possible combination of agent i in a multi-agent system, Indicates that agent i is in the combination The marginal utility generated in .

4. The method for testing autonomous driving based on multi-agent cluster confrontation according to claim 1, characterized in that: The strength of the confrontation Adaptive dynamic update using reverse update algorithm, the update expression is: , Where: for The intensity of the confrontation at all times, Indicates the update rate of the adversarial strength Estimated effect of the tested autonomous driving vehicle facing cluster confrontation.

5. The method for testing autonomous driving based on multi-agent cluster confrontation according to claim 1, characterized in that: In step S3, the minimum sequence quadratic programming method is used to solve the optimal Nash equilibrium confrontation strategy of the multi-agent system against the tested autonomous driving vehicle.

6. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images