MPPF: Mobile Probe Potential Field for Real-Time Multi-Agent Navigation in Industrial Environments

Industrial inspection in open outdoor unstructured multi-terrain environments creates critical autonomous navigation challenges. Ground agents required for manipulation and close inspection cannot achieve the comprehensive environmental perception necessary for safe autonomous operation, necessitating multi-agent collaboration. This necessitates parallel processing of collaborative tasks, including trajectory planning, perception, real-time dynamic obstacle detection, and avoidance on resource-constrained edge devices. Terrain variations further demand adaptive locomotion strategies, necessitating legged robot deployment with adaptive gait patterns. To address these challenges, we propose a collaborative multi-agent framework where a UAV provides environmental perception, localization, and dynamic obstacle detection through RGB-D cameras, utilizing deep learning for RGB processing and classical vision for depth analysis. The quadruped receives real-time guidance while executing terrain-adaptive locomotion with adaptive gait switching. For efficient trajectory planning under computational constraints, we present MPPF (Mobile Probe Potential Field), a moving surface charge model for obstacle representation in APF methods. Our algorithm incorporates a virtual obstacle-based framework for local minima resolution and smooth trajectory generation through dynamic parameter adaptation. Experimental validation of our proposed framework demonstrates trajectory generation in 0.025 seconds with superior smoothness compared to baseline methods, achieving successful navigation across sand, stairs, and stones for autonomous industrial inspection in unstructured environments.