车辆自主决策安全性验证方法

A Security Verification Approach for Autonomous Maneuver Decision of Vehicles

针对传统算法在车辆自主决策安全性验证中置信水平较低、验证结果不完备的问题,基于形式化的可达性方法分析车辆自主决策行为状态的可行性,并给出车辆自主决策的行为状态完备性集合。采用有向凸多面体算法对自主车辆非线性混合系统进行保守线性化处理,再利用迭代原理和Zonotope近似方法计算可达集。变道避碰的仿真实验表明,在初始状态集和传感器噪声边界条件下,自主车辆位置可达集和速度可达集给出了高置信水平的安全保证。

In order to sovel the problem that the traditional algorithm has low confidence level and incomplete verification results in vehicle autonomy decision security verification,the formalized accessibility method was used to investigate the feasibility of autonomous decision-making behavior state of vehicles and gives the complete set of autonomous decision-making behavior state of vehicles.The directed convex polyhedron algorithm is used to conservatively linearized the nonlinear hybrid system of autonomous vehicles,and the reachable set is calculated by iterative principle and Zonotope approximation method.Simulation results of lane change collision avoidance show that under the initial state set and sensor noise boundary conditions,the position accessible set and speed accessible set of autonomous vehicles provide a high level of confidence for safety.