Path Following with Supervised Deep Reinforcement Learning
Wen-Yi Gu, Xin Xu, Jian Yang · 2017
In this paper, we propose a fast adaptive learning method called supervised deep reinforcement learning to realize path following with high-dimensional input for autonomous driving task. We combine traditional feedback control method with deep reinforcement learning, the former providing a basic steering manipulation technique and the latter further improving the performance, which is similar with human's learning process. We validate our approach on computer simulating driving task. Experiments show that the fusion method steadily improves the performance based on the result of feedback control.