Research on Optimization of Single Vehicle Trajectory Planning Based on Heuristic Algorithm

In this paper, the problem of single-vehicle trajectory planning in self-driving cars is studied in an in-depth way, aiming at improving the safety and efficiency of self-driving by modeling the Optimal Control Problem, OCP). First of all, the development background of autonomous driving technology and its application importance in automobile industry are introduced, and then the components of automobile autonomous driving architecture are elaborated in detail, with special emphasis on the key role of trajectory planning. In this paper, OCP problem is transformed into Nonlinear Programming, NLP) problem and solved by gradient descent method. At the same time, it is pointed out that the determination of initial solution is important for solving the optimal control problem of path planning. In terms of path planning methods, this paper discusses five mainstream methods: graph theory, RRT, reinforcement learning, heuristic algorithm and optimization algorithm. Each method has its advantages and limitations, such as the efficiency of $A^{*}$ algorithm, the convergence speed of RRT algorithm, and the inexplicability of reinforcement learning. In order to solve these problems, this paper proposes an initial solution design method based on genetic algorithm, including particle swarm optimization (PSO) and sine and cosine algorithm (SCA), and compares the performances of these two algorithms through simulation experiments. The experimental results show that SCA algorithm is superior to PSO algorithm in convergence time, and PSO algorithm performs better in solving quality. Finally, this paper summarizes the importance of modeling the single-vehicle trajectory planning system, and points out that the initial solution design based on genetic algorithm has a significant effect on solving OCP problems.