Research on Continuous Control Approach of Mobile Robots for Mapless Navigation
Qiang Zou, Jing-Yuan Liu, Ming Cong, Dong Liu, Guiyuan Wang · 2024
Mapless navigation is a widely-used technique for directing mobile robots through an unknown environment. Inspired by goal-oriented navigation features of mammals in nature, this work introduces heuristic knowledge and proposes a method known as Heuristic Knowledge based Deep Deterministic Policy Gradient, HK-DDPG, to direct the robot's motion actions, which can reduce the randomness of actions effectively. To solve the issues of low sample date utilization and poor training stability in DDPG, HK-DDPG incorporates a temporal difference error and a weight-based optimization sampling mechanism. The results of simulation and real environment evaluation experiments demonstrate that HK-DDPG algorithm exhibits faster convergence, higher learning efficiency, and better stability than DDPG algorithm. The robot deployed with the proposed model can be extended for application in real-world scenarios.