Flex-π: Multi-Stream World-Action Model with Compute Flexibility
chris_j_paxton · x · 2026-08-13
Flex-π is a 6B-parameter world-action model (WAM) that jointly predicts RGB, 3D geometry, and DINO semantics, enabling flexible deployment from action-only to full joint generation. It outperforms strongest baselines by 2-7x on dexterous tasks and runs faster than π0.5. Paper from UW and AI2.
Related event: Flex-π: 6B Parameter Robotic Model Released(2 posts)→
More from Embodied
- Inside Unitree's IPO Roadshow: 70% Revenue from Research, Plans for US Plant — MatthewChang · 2026-08-13
- Sandbar Introduces Stream, a Push-to-Talk AI Ring — RebeccaBellan · 2026-08-13
- Deep Dive with Unitree Founder: Details Behind the G1 Humanoid Robot — MarwaEldiwiny · 2026-08-13
- Robotics Artificial Muscles Testing: Developer Measures Actuation Speed — IanPritchard · 2026-08-13
- Embodied AI Breakthrough: SONIC System for Robot Motion Tracking Published in Science Robotics — zhengyiluo · 2026-08-13
- Musk: Digital Optimus to replace all office jobs, Optimus all physical labor — davidpattersonx · 2026-08-13