Near-Optimal Synchronization of Multiple Robot Carts Using Online Reinforcement Learning
Jakob Harig, Ryan Russell, Jing Wang, Yufeng Lu · 2022
This paper studies the optimal synchronization of multiple robot carts using online reinforcement learning. The objective is to validate the use of online reinforcement learning to cooperatively control multiagent systems, which may render the potential applications in optimization and control of engineered multiple dynamical systems, such as autonomous vehicles platoon, power grids, sensor networks, et.al. To control the robots, we pro-pose an online reinforcement learning based cooperative control algorithm, which enables the individual agent to learn its own optimal control law in real time by minimizing the overall group cost function. The result of the online reinforcement learning leads to a model free near optimal cooperative control algorithm for multiagent systems. To experimentally verify the control algorithm, two robots are constructed with several integrated sensor modules. Via the XBee module, robots can communicate position and velocity information to one another and receive position and velocity information from a virtual leader. This virtual leader is following a desired trajectory, allowing us to see the control algorithm synchronize the two followers with the leader on a desired trajectory. The control algorithm is implemented using an NVIDIA Jetson Nano. The testing results have proved the effectiveness of the proposed design.