This study introduces a method for multi-robot motion planning that combines centralized, sampling-based tree search with decentralized, data-driven steer and distance heuristics. The method has been tested on a range of robot and obstacle densities, showing its ability to plan for up to 16 robots and validating the effectiveness of data-driven heuristics in combating exponential search space growth. The research extends the use of data-driven heuristics in motion planning to multi-robot systems and demonstrates the successful decomposition of high-dimensional joint-space motion planning problems into local problems.
Publication date: 22 Nov 2023
Project Page: https://www.caltech.edu/
Paper: https://arxiv.org/pdf/2311.12385
