Deep Q-Network based Tracking and Slope Climbing Controller Design of Mobile Robot and Flatbed Vehicle

Thanks to the technological change, the industry 4.0 has arrived. Artificial intelligence let software and machines to sense and learn on their own activities. In these research field, formation control of mobile robots is a popular topic. From rescue task to environment exploration, many of the applications require the robots to perform group patterns. To achieve these works, the control process could be simplified into multiple leader-follower relationships which control the followers to achieve the specific relative position with respect to the leaders. However, in particular situations, the tracking process should further consider the relative pose when both robots reach the desired relative position such as dynamic combination of robots. This paper proposes a Deep Q-Network based leader-follower tracking method with a climbing-up combination process. The control algorithm depends on the robot vision of the follower without any global measurements, and the whole process focuses on the relative pose between the mobile robot (follower) and the flatbed vehicle (leader).