GPU-accelerated Incremental Euclidean Distance Transform for Online Motion Planning of Mobile Robots

This research letter introduces a volumetric mapping system designed for mobile robots that utilizes parallel computing via GPU programming to efficiently calculate Occupancy Grid Maps (OGMs) and Euclidean Distance Transforms (EDTs). Unlike systems focused on high-precision structural reconstruction, this approach incrementally builds a global EDT while outputting high-frequency local distance information specifically for online motion planning. The system integrates continuous observations into the global EDT using a parallel wavefront algorithm while preserving historical data. Experimental results indicate that the proposed method outperforms existing state-of-the-art robot mapping systems, particularly in mapping unexplored areas, and achieves real-time performance on both aerial and ground vehicles despite limited onboard computational resources.