Federated Reinforcement Learning Acceleration Method for Precise Control of Multiple Devices

Nowadays, Reinforcement Learning (RL) is applied to various real-world tasks and attracts much attention in the fields of games, robotics, autonomous driving. It very challenging devices overwhelming directly apply RL environments. Due reality gap simulated environment does not match perfectly scenario additional learning cannot be performed. Therefore, an efficient approach required for find optimal control policy get better efficacy. In this paper, we propose federated reinforcement based on multi agent which applying a new federation policy. The allows agents perform share their experiences with each other e.g., gradient model parameters increase level. Actor-Critic PPO algorithm used four types simulation environments, OpenAI Gym's CartPole, MoutainCar, Acrobot, Pendulum. addition, did real experiments multiple Rotary Inverted Pendulum (RIP) evaluate compare efficiency proposed scheme both