CASIA's PhiZero learns 'physical language' to lead three generation benchmarks without natural language descriptions
jiqizhixin · x · 2026-08-25
The Institute of Automation, Chinese Academy of Sciences (CASIA) presents PhiZero, a world model that learns a "physical language" instead of relying on natural language descriptions of physics.
Core Innovation:
- It learns a compact symbolic system that directly represents fine-grained state changes in the real world.
- The model predicts state evolution in this learned language and decodes it into video or transfers state transitions across different embodiments.
Performance:
- Generation: Leads on Physics-IQ Verified, PhyGround, and WorldModelBench.
- Understanding: Posts competitive results on IntPhys2, LikePhys, and YoCausal.
- Zero-shot Transfer: Successfully transfers motion from humans to G1 humanoids, from human hands to dexterous grippers, and from simulation to reality without paired training data.
More from Embodied
- Robotics LLMs tested on 16 tasks: GPT-6 Astra hits 17.2% success vs Fable's 9.1% — pabbeel · 2026-10-16
- WEDGE paper generates training environments conditioned on navigation agent failures — _akhaliq · 2026-10-16
- Viral meme nails AI's irony: it can write symphonies but can't do the dishes — briannekimmel · 2026-10-16
- Knödel Watch: animated vocabulary cards on ESP32-S3, open-sourced 4 days after unboxing — alexcovo_eth · 2026-10-16
- NVIDIA RTX Spark PC launches tomorrow with global events and strong early reviews — NVIDIA Blog · 2026-10-16
- SlipSense fuses pressure and vibration for 23ms zero-shot robotic slip detection — micoolcho · 2026-10-16