| CPC G06N 7/01 (2023.01) [G06N 3/006 (2013.01); B60K 2360/175 (2024.01); B60Q 2800/10 (2022.05); G05B 2219/39146 (2013.01); G06N 3/092 (2023.01); G06N 3/094 (2023.01); G06N 20/00 (2019.01)] | 15 Claims |
|
1. An autonomous driving test method based on multi-agent swarm adversarial, comprising steps of:
step S1: deducing a conflict topological relationship graph between a tested autonomous vehicle and an agent according to a road topology of a test scenario and a conflict relationship of test objects, specifically comprises:
deducing whether a spatial conflict exists between the tested autonomous vehicle and a multi-agent in an environment, and among multi-agents, based on vehicle state information of a multi-agent vehicle group {Veh1, Veh2 . . . Vehn} and the tested autonomous vehicle Veh0, and test map information; and
adopting graph theory to describe a topological relationship of a vehicle conflict to obtain conflict topological relationship graph G=(V,E) of the tested autonomous vehicle and the agent, wherein V represents a set of vehicles, vehicle Vehn had a position of pn=(xn(t),yn(t)) and a speed of vn(t) at time t; and E represents a set of edges, for edge eij, an inference is made according to current position pi of vehicle Vehi and current position pj of vehicle Vehj, and if a spatial conflict exists, then the edge is recorded as eij=1, and otherwise, the edge is recorded as 0;
step S2: deducing a feasible planning space of the tested autonomous vehicle according to the conflict topological relationship graph;
step S3: establishing a multi-agent swarm adversarial model based on a potential game under the feasible planning space according to a correlation between an individual reward of the agent and a swarm adversarial test effect of a multi-agent system, and solving and obtaining an optimal adversarial strategy of the multi-agent system against the tested autonomous vehicle, wherein in the multi-agent swarm adversarial model, an adversarial intensity is introduced to characterize relative weights of the individual reward of the agent and the swarm adversarial test effect of the multi-agent system, and the adversarial intensity is adaptively adjusted according to an actual response of the tested autonomous vehicle;
for the establishing a multi-agent swarm adversarial model based on a potential game under the feasible planning space according to a correlation between the individual reward of the agent and a swarm adversarial test effect of the multi-agent system, an expression is:
 wherein in the expression: P(aix, pi, p0) represents a swarm adversarial effect of a multi-agent system of agent i when an adversarial strategy is aix, and U is the feasible planning space; P(ai0, pi, p0) represents a swarm adversarial effect of the multi-agent system of the agent i under any initial adversarial strategy; Ri(aix, pi, p0) represents an individual reward of the agent i when an adversarial strategy is aix, and Ri(ai0, pi, p0) represents an individual reward of the agent i under an initial adversarial strategy; and ai is an acceleration of the agent i, pi is a position of the agent i, and p0 is a position of the tested autonomous vehicle;
for the individual reward of the agent, a function expression is:
 wherein in the expression: rself,it(ai, pi) represents a driving reward of the agent i at the time t, ai is the acceleration of the agent i, pi is the position of the agent i, ddes,it is a distance between the agent i and an end point, and jit is a jerk of the agent i; rgroup,i0t(ai, pi, p0) represents an adversarial reward of the agent i at the time t, ΔTTCPi0t represents a time difference between the agent i and the tested autonomous vehicle Veh0 reaching a conflict point at the time t, dcp,it represents a distance between the agent i and the conflict point, vit represents a speed of the agent i, dcp,0t represents a distance between the tested autonomous vehicle and the conflict point, v0t represents a speed of the tested autonomous vehicle, and p0 is the position of the tested autonomous vehicle; and θ is the adversarial intensity to characterize relative weights of the individual reward of the agent and the swarm adversarial test effect of the multi-agent system; and
for the swarm adversarial test effect of the multi-agent system, a function expression is:
 wherein in the expression: γ represents a reward reduction coefficient; T is a planning step size; and φi is a contribution generated by the agent i in adversarial; and
step S4: repeatedly executing the steps S1-S3 until an adversarial task is completed.
|