Semantic potential field for mobile robot navigation using grid maps

Abstract

Traditional navigation methods for mobile robots face significant challenges in dynamic environments, including local minima avoidance and efficient path planning. This paper introduces the semantic potential field (SPF) method, which synergizes geometric and semantic data using a semantic grid map to improve navigation efficiency and adaptability. The key features of the SPF method include (i) a semantic grid map combining light detection and ranging (LiDAR) and camera data to distinguish static and dynamic obstacles and (ii) a dynamically modulated potential field incorporating semantic weights for adaptive path planning and obstacle avoidance. The experimental results demonstrate that the SPF method significantly reduces the travel distance and computation time compared with those of traditional methods, ensuring robust navigation in diverse environments. By addressing the limitations in real-time navigation systems, the SPF represents a significant advancement in mobile robot path planning, with promising applications in disaster response, autonomous logistics, and defense.