Navigating the Clutter: Waypoint-Based Bi-Level Planning for Multi-Robot Systems

Researchers from MIT, Google, and other institutions have developed a waypoint-based bi-level planning framework for multi-robot systems operating in cluttered environments. The approach addresses the challenge of joint optimization across high-level task planning and low-level motion planning, which is difficult due to complex parameterization of motion trajectories and credit assignment ambiguity. By introducing waypoints as a simple yet expressive representation for motion trajectories, the framework enables effective parameterization of low-level planning while using a curriculum-based training strategy with a modified RLVR algorithm to propagate motion feasibility feedback to the task planner.

Experiments on the BoxNet3D-OBS benchmark, which features dense obstacles and up to nine robots, show consistent improvements in task success over motion-agnostic and VLA-based baselines. The code is publicly available, making the approach accessible for further research and real-world applications in warehouse automation, search-and-rescue, and other domains requiring coordinated multi-robot navigation.