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

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