Study on Deep Reinforcement Learning for Mobile Robots Flocking Control in Certainty Situations

摘要

Mobile robots are the widely used machines that will be a part of life. It can easily be scaled to meet the requirements of human uses and can move to different target areas freely. However, mobile robots have a problem with mobility, which needs to respond to human activities and obstacles. Reinforcement learning (RT) is a part of machine learning, which enables the machine to learn by themselves. It can develop in a real-world environment without machine teaching or patterns. This paper proposed the study on the flocking control simulation, which avoids the obstacles. Mobile robots in simulation presented by using Multi-Agent Deep Deterministic Policy Gradient (MAD-DPG). The proposed algorithm, the deep reinforcement learning algorithm, is the main navigation of robots, which find the target, maintain distance between robots, and avoid collision with obstacles by learning features and characteristics with certainty environment, obstacle, robots, and target. On the simulation, the authors operate all of the positions in the environment, and robots speed in the environment as the state.