Mobile Robot Planner with Low-cost Cameras Using Deep Reinforcement Learning

摘要

This study develops a robot mobility policy based on deep reinforcement learning. Since traditional methods of conventional robotic navigation depend on accurate map reproduction as well as require high-end sensors, learning-based methods are positive trends, especially deep reinforcement learning. The problem is modeled in the form of a Markov Decision Process (MDP) with the agent being a mobile robot. Its state of view is obtained by the input sensors such as laser findings or cameras and the purpose is navigating to the goal without any collision. There have been many deep learning methods that solve this problem. However, in order to bring robots to market, low-cost mass production is also an issue that needs to be addressed. Therefore, this work attempts to construct a pseudo laser findings system based on direct depth matrix prediction from a single camera image while still retaining stable performances. Experiment results show that they are directly comparable with others using high-priced sensors.