A Dueling-DDPG Architecture for Mobile Robots Path Planning Based on Laser Range Findings

Panpan Zhao, Jinfang Zheng, Qinglin Zhou, Chen Lyu, Lei Lyu

Abstract: Planning an obstacle-free optimal path presents great challenges for mobile robot applications, the deep deterministic policy gradient (DDPG) algorithm offers an effective solution. However, when the original DDPG is applied to robot path planning, there remains many problems such as inefficient learning and slow convergence that can adversely affect the ability to acquire optimal path. In response to these concerns, we propose an innovative framework named dueling deep deterministic policy gradient (D-DDPG) in this paper. First of all, we integrate the dueling network into the critic network to improve the estimation accuracy of Q-value. Furthermore, we design a novel reward function by combining the cosine distance with the Euclidean distance to improve learning efficiency. Our proposed model is validated by several experiments conducted in the simulation platform Gazebo. Experiments results demonstrate that our proposed model has the better path planning capability even in the unknown environment.