Pose Control of Omnidirectional Mobile Robot Using Improved Deep Reinforcement Learning

Nowadays, mobile robots are being widely applied in various fields such as indoor carrying and check of products and outdoor exploration. One of the most important problems arising in development of mobile robots is to resolve path planning problem. With active studies of implementation of path planning, lots of algorithms have been developed and especially, the dramatic advance in artificial intelligence (AI) led to advent of algorithms using reinforcement learning (RL). Deep reinforcement learning (DRL) has been developed and it uses neural network to approximate parameters of RL algorithm. DDPG is one of deep reinforcement learning (RL) algorithms and is widely used to solve lots of practical issues as it doesn’t need full information of the environment. In other words, path planning with DRL has advantages of possibility for unknown environments in which partial or full information is not given and of direct controllability of the robot. Generally, path planning Up to now, path planning using DRL has considered only position control problem with no consideration of its orientation angle (as the author knows). In this paper, a pose control method using DRL for 3-wheeled omnidirectional mobile robot is proposed. And a method to reduce position error is mentioned. Simulation results show that the proposed method can efficiently solve the control problem of omnidirectional robots.