Local Path Planning with Turnabouts for Mobile Robot by Deep Deterministic Policy Gradient

This paper proposes local path planning with turnabouts for a mobile robot by deep deterministic policy gradient (DDPG). DDPG is one of the actual reinforcement learning methods. This paper focuses on a non-holonomic mobile robot that has a minimum turning radius. Narrow roads exist in human living areas such as homes, commercial facilities, and factories. In this paper, a narrow road is defined as an impassable road with the minimum turning radius of the robot. Therefore, local path planning with turnabouts is needed for a mobile robot to pass a narrow road. However, most conventional local path planning methods do not consider turnabouts, since these methods select only forward velocity. This paper generates the local path planning which consists of forward and backward motion by using DDPG. For the trained model, simulation is used to obtain optimal velocity by minimizing the long-term reward. The reward is set considering goal arrival, number of turnabouts, and obstacle avoidance. The validity of the proposed local path planning by DDPG was confirmed by simulation and experimental results.