Denoising Multi-Robot Trajectories

Abstract

A family of multi-robot trajectory planners built on D4orm, a dynamics-aware diffusion-denoising framework that refines candidate control trajectories through massively parallel, GPU-friendly sampling. Decoupled, online receding-horizon, and distributed variants target scalability, feedback control, and resource-constrained settings. Demonstrated with zero-shot deployment on ten real quadrotors, deconfliction of 100 simulated robots, and fully onboard distributed operation on six ground robots.

Paper