Mirror video planner shows emergent in-context learning, boosting robot lava-lake escape rate from 46% to 76%
vincesitzmann · x · 2026-10-20
Mirror, a video-model-based planner, shows early signs of emergent behavior: it learns from its own history in-context without any finetuning.
- Dropped into a lava lake, it reaches a 76% cumulative escape rate over 30 attempts, vs. 46% when context is reset after every try — it remembers failed strategies and biases future sampling toward successful ones.
- The authors stress Mirror is not RL nor classic behavior cloning: it needs no video-action pairs, but is a new approach to learning-from-experience built on video models and LLMs.
- Vince Sitzmann predicts a revival of the "arm-farm" paradigm for improving robot policies, arguing it will work this time because LLMs and video models make exploration sampling scalable.
More from Embodied
- "Robots are going underground, figuring out their own routes" — a look past chatbot debates — VraserX · 2026-10-21
- Deepak Pathak: There Is No Perfect Data Source for Robotics — chris_j_paxton · 2026-10-21
- Discrete action representations match or beat continuous ones for robot policies — GuanyaShi · 2026-10-21
- Developer gave his Grok bot $1,000 to build itself a body — parts keep arriving — Baconbrix · 2026-10-21
- Tuesday Labs opens $200 reservations for 999 numbered Voxel Series-1 home robots — testingcatalog · 2026-10-21
- Tuesday Labs' home robot Voxel charms with Pixar-collaborated personality design — cneuralnetwork · 2026-10-20