Local Trajectory Planning of Mobile Robot with Deep Reinforcement Learning Based on Q Value

摘要

The deep reinforcement learning algorithm based on visual perception and intelligent decision combines the perception ability of convolutional neural network with the decision control ability of reinforcement learning via end-toend learning style and realizes the process from raw visual input to decision action output. It has been extensively applied to high-dimensional visual input and decision control tasks since it was put forward. In this paper, the deep reinforcement learning algorithm based on Q value was proposed to realize local trajectory planning of mobile robot in a dynamic environment. Compared with the vulnerability of artificial design expert system, this algorithm possesses stronger robustness. By realizing the transformation from experience-driven man-made features into data-driven representation learning, this algorithm has greatly improved the real-time obstacle avoidance performance of robots.