Safe and trustworthy robot pathfinding with BIM, MHA*, and NLP

Abstract

Abstract Construction robotics has gained significant traction in research and development, yet deploying robots in construction environments presents unique challenges. Construction sites are characterized by indeterminate processes, domain-specific tasks, and complex navigation requirements that make traditional robot pathfinding approaches insufficient. While methods like simultaneous localization and mapping (SLAM) offer viable solutions for robot navigation, they often require considerable computational resources due to their sensor precision demands and data processing needs. In the context of construction robotics, building information modeling (BIM) has emerged as the leading map representation for robot planning. However, the assumption of having a perfectly accurate model is dangerously flawed, and thus algorithms such as A* do not constitute a reliable approach for general robot planning. We integrate a multi-heuristic A* (MHA*) algorithm with artificial potential fields (APF) derived from BIM spatial data, while employing large language models (LLMs) to process BIM’s textual information for dynamic obstacle avoidance. Our experimental results demonstrate an 80% improvement in robot-obstacle clearance while maintaining comparable path lengths to traditional methods. This approach provides a computationally efficient solution for safe robot navigation in complex construction environments.