Path Planning Method for Autonomous Driving Vehicles Based on Deep Convolutional Neural Network

Engineering notes

Key topics: autonomous driving, autonomous vehicle, path planning, reinforcement learning, carla, perception, planning. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

This paper proposes a dynamic adaptive convolutional path planning algorithm for path planning of autonomous vehicles. The algorithm constructs an environment perception module with multi-scale feature fusion, combines the dynamic adaptive convolution module with the reinforcement learning path decision mechanism, and realizes accurate perception of complex road environments and efficient path planning. On the Carla simulation platform, based on the public autonomous driving data set, simple roads, complex urban roads and highways are set up for experiments. The path length, planning time and collision rate are used as evaluation indicators to compare with $\mathrm{A}^{*}$, Dijkstra and traditional convolutional neural network path planning algorithms. The experimental results show that in complex urban road scenarios, the average path length of the proposed algorithm is shortened by 12.3% compared with the $\mathrm{A}^{*}$ algorithm, the planning time is reduced by 18.7%, and the collision rate is as low as 0.03%. Compared with the traditional convolutional neural network path planning algorithm, the collision rate is reduced by 21.6%; in the highway scenario, the average path length is optimized by 8.9%, and the planning time is shortened by 15.2%. The experimental data show that the algorithm has significant advantages in path planning accuracy, real-time and safety, and effectively improves the path planning performance of autonomous vehicles.

Links and sources

• Official / arXiv page

Comments