Product · mentioned 6 times
An imitation-learning algorithm for robot manipulation that captures multiple modes of behaviour for the same observation while keeping training stable.
No connections recorded yet.
RoboPapers· 10 Dec 2025
Zhenyu Zhao says they were surprised that DP3 worked better than diffusion policy, and that using depth as an observation works better than raw RGB images.No Priors: Artificial Intelligence | Technology | Startups· 19 Nov 2025
Cheng Chi says diffusion policy unlocked more scalable training and data collection, so multiple people, sometimes untrained, can collect data and still get good results.