Narrow-Route Path Planning for Mobile Robots Using Deep Deterministic Policy Gradient Considering Turning Radius Limit

This paper proposes a narrow-route path planning for a mobile robot using deep deterministic policy gradient (DDPG) considering a drive system. In this paper, a narrow road is defined as a space in which a mobile robot with a non-holonomic constraint cannot move without performing turnabouts. There are various drive systems for mobile robots such as an independent two-driven wheels type, a steering type, and a car-like type. In an independent two-driven wheels type, the mobile robot plans the path including on-the-spot turning. On the other hand, in a steering type and a car-like type, the mobile robot performs the turnabouts on a narrow road. In wheeled robots, differences in drive systems can be expressed as a turning radius limit. The proposed method generates narrow-route path planning considering a turning radius limit due to a drive system. The proposed method is based on machine learning and uses DDPG as reinforcement learning. The trained model determines the translational and angular velocities that include turnabouts / on-the-spot turning according to environmental information in real time. In the simulation and experiments, we confirmed that the proposed method allowed a mobile robot, with or without a turning radius limit, to pass through a narrow road. In addition, the robustness against the trained model was evaluated by several narrow roads that differed from the learning environment. In the case of the drive system with the turning radius limit, the success rate of driving on narrow roads including different learning environments was 94% in simulation and 85% in experiments. Therefore, the effectiveness of the proposed method was confirmed by the simulation and experimental results.