THREAD: Trajectory Planning for Hybrid Rigid-Soft Manipulators with Environment-Aware Diffusion

THREAD is a novel diffusion-based trajectory planner designed for hybrid rigid-soft manipulators, addressing challenges in confined environments by learning generative priors over feasible backbone trajectories conditioned on local geometry. It incorporates physics-inspired losses to enforce curvature, smoothness, and collision constraints across both rigid and soft segments. Achieving a 92.4% task success rate with five times fewer collisions than the strongest baseline, THREAD demonstrates effective cross-embodiment transfer to real-world applications, enabling successful manipulation through narrow apertures.