HT-Policy: multimodality isn't why diffusion policies win at imitation learning
chris_j_paxton · x · 2026-11-18
The common explanation for diffusion/flow-matching policies beating regression in robot imitation learning—better support for multimodality—may be wrong, argue researchers Yuchen and Maojiaoyuan. Real-world success actually hinges on stripping multimodality from robot data. They attribute the advantage to state-dependent variation and heavier-than-usual Gaussian tails, and propose HT-Policy, which learns an input-dependent residual scale to mitigate long-tail samples, making plain regression competitive with flow matching while being faster and simpler to train. Chris Paxton highlights the accompanying RoboPapers podcast on robot learning theory.
More from Embodied
- UMA Robotics bets on Europe first: survey shows EU edges US on talent, customers, cost — RemiCadene · 2026-11-18
- Tesla letter: FSD hits 1.74M paid users, Robotaxi in 7 cities ahead of Dec 3 vote — yunta_tsai · 2026-11-18
- Open-Source Robot Lamp Preps Skill Store: Install Physical Skills Like Apps — dee_hw · 2026-11-17
- Genesis AI raises $500M Series A at $3.5B valuation for general-purpose robots — chris_j_paxton · 2026-11-17
- Robotics researchers spar: pick-and-place with dexterous hands is 'so boring' — chris_j_paxton · 2026-11-17
- Open-Source reBot Arm B601-DM Robotic Arm on Sale for $1,090 — philfung · 2026-11-17