A new robot imitation method learns 1,000 tasks in under 24 hours of demo time
chris_j_paxton · x · 2026-07-30
A Science Robotics paper reports a new imitation-learning method that taught a robot 1,000 distinct everyday manipulation tasks in under 24 hours of human demonstrator time.
The work studies two priors for data efficiency: splitting manipulation into alignment and interaction phases, and using retrieval-based generalization. Across 3,450 real-world rollouts, the authors found that this decomposition gave an order-of-magnitude gain in the few-demonstrations-per-task regime, with retrieval consistently beating monolithic behavioral cloning. The resulting method, MT3, can learn some tasks from a single demo and generalize to unseen object instances.
The paper also notes limits: the gains depend on the task family, and the method was evaluated on real-world rollouts rather than simulated toy settings.
More from Embodied
- Ray-Ban Meta Sales Approach 10M Units: AI Eyewear Reshapes the Consumer Electronics Ecosystem — Annual_Judge_7272 · 2026-07-30
- One Hacker Controlled 7,000 Home Robots: US Regulation Targets Networked Device Threats — Illustrious-Data1712 · 2026-07-30
- A former RRE investor breaks down who could win the humanoid robotics race — MarwaEldiwiny · 2026-07-30
- NVIDIA pitches Jetson AGX Orin’s 275 TOPS for robots and edge AI — NVIDIA Developer · 2026-07-30
- Two-wheeled biped robot shows why deployment matters more than raw speed — Olivier__OG · 2026-07-30
- US drone factory says it has closed fresh funding for high-performance AI drones — omojumiller · 2026-07-30