A Knowledge-based Full-time Non-homotopy Path Optimization Method Through Online Environmental Learning

Recently, many researchers propose different methods to solve the motion planning problem in the human-robot co-existing environments. However, when encountering crowds that suddenly appear, conventional motion planning algorithms often generate similar homotopy paths with slight local variations, which may not generate new feasible paths in time and thus lead to congestion for the robot. In order to address this problem, this paper proposes a knowledge-based full-time non-homotopy path optimization method through online environmental learning, which generates heuristic paths to guide motion planning by combing the pedestrian knowledge and the environmental structure. Firstly, a pedestrian matrix based knowledge base is proposed to record the pedestrian flow pattern with respect to different locations, and the corresponding pedestrian matrix is extracted according to the perception during the robot navigation process. Secondly, the environmental perception is performed surrounding the key obstacles that appears around the crowds to extract the feature points. Then, the extracted feature points are reused as prior knowledge in subsequent navigation. When crowds suddenly appear in the global path, the algorithm will prioritize searching a non-homotopy path to avoid crowds. The experimental results show that the method proposed in this paper can significantly reduce the time required for the robot to navigate and improve the success rate of navigation in crowded environments.