基于凸近似避障原理的自动驾驶集成决策与控制方法

Intelligent Transportation Integrated Decision and Control Method for Autonomous Driving Based on Convex Approximation Obstacle Avoidance Principle

针对结构化道路下自动驾驶汽车的轨迹跟踪问题,提出了一种基于凸近似避障原理的自动驾驶集成决策与控制方法。首先基于凸近似避障原理,对安全约束进行优化,适当缩小轨迹可行域,只保留与特定周车交互相关的部分可行点;然后结合模型预测控制算法,建立低速场景下线性化的自行车运动学模型,以轨迹跟踪误差最小为目标,考虑自车和周车的外形、道路几何约束和安全约束,构造多个与静态路径相关的最优控制问题,使用外罚函数处理约束,基于序列二次规划方法进行求解,选择最优轨迹进行跟踪。在Carla仿真平台上的高速公路仿真实验结果表明提出的基于凸近似避障原理的自动驾驶集成决策与控制方法虽然通过效率有所下降,但是驾驶安全性得到充分保障,并且轨迹跟踪性能未受影响,因此该方法能够有效降低自动驾驶中的决策风险。

Aiming at the trajectory tracking problem of self-driving cars under structured roads,an integrated decision-making and control method for self-driving based on the convex approximation obstacle avoidance principle is proposed.First,based on the convex approximation obstacle avoidance principle,the safety constraints are optimized,the feasible trajectory domain is appropriately reduced,and only the part of the feasible points related to the interaction with a specific week vehicle is retained.Then,in combination with the model predictive control algorithm,a linearized bicycle kinematic model is established in the low-speed scenario,and with the goal of minimizing the trajectory tracking error,the shapes of the self-driving car and the weekly vehicle,the road geometric constraints,and the safety constraints are considered,and multiple optimal control problems are constructed in relation to the static paths.The optimal control problems in relation to the static paths are constructed,and the constraints are handled using an external penalty function that is solved based on the sequential quadratic programming method to select the optimal trajectory for tracking.The results of highway simulation experiments on the Carla simulation platform show that the proposed integrated decision control method for autonomous driving based on the convex approximation obstacle avoidance principle can effectively reduce the decision risk in autonomous driving,although the overtaking efficiency has been reduced,but the driving safety has been adequately guaranteed and the trajectory tracking performance has not been affected.