Berkeley trains a diffusion model on RL policy weights to skill a real G1
FinanceYF5 · x · 2026-10-26
Berkeley researchers trained a diffusion model on weights from thousands of RL policies; its generated policies enable a real Unitree G1 to walk, dance, and do spinning kicks — generating robot skills like images.
Also featured: KAIST's Humanoid v0.9 uses onboard vision to push up to 35 kg and lift 20 kg, capping the demo with a moonwalk.
Related event: Berkeley Generates RL Policies with Diffusion Models for Real Robot Skills(2 posts)→
More from Embodied
- OmniEgoCap Turns Any Egocentric Device into a Full-Body 3D Mocap System — Scobleizer · 2026-10-26
- Tesla FSD approved in Latvia, the 10th European country — mitchdeg · 2026-10-26
- On-device VLM goes production: GenieX runs Liquid AI's LFM-VL live at PyTorch Conference — PyTorch · 2026-10-26
- One long Claude Code session builds a full Android TV launcher with a 1-second-latency live camera — aristofeles · 2026-10-26
- LeCun co-authored Temporal Straightening: straightening latent paths boosts gradient-based planning success rates — ylecun · 2026-10-26
- Humanoid robots' DRAM demand may 20x, but it's still just 0.04% of 2026 supply — zephyr_z9 · 2026-10-26