Paper 2607.28624
PhiZero: A World Model Built Around Physical Language
- Published
- Jul 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- 77 stars
01 In brief
Summary
PHIZERO is a physical world model that learns a compact discrete 'physical language' from in-the-wild videos via self-supervision, representing state transitions.
It adopts a reason-then-render paradigm: a Physical Language Reasoner (initialized from Qwen3-VL-4B) predicts a physical-language sequence from the first frame and a textual action intent, and a diffusion decoder (based on Wan2.2-5B) renders the inferred transitions into video.
The Physical Language Tokenizer uses a transition-level Q-Former and FSQ to compress video into 256 discrete symbols per 33-frame clip, with a pure-noise warm-up to avoid shortcuts.
Training uses a hierarchical data pipeline (50K hours real, 1K hours simulation) and curriculum.
Experiments show PHIZERO outperforms baselines on generation benchmarks (Physics-IQ Verified, PhyGround, WorldModelBench) and understanding benchmarks (IntPhys2, LikePhys, YoCausal).
It also enables zero-shot motion transfer across embodiments (human to robot) and sim-to-real transfer, and supports action-conditioned and interactive world modeling.
Ablations confirm the importance of the diffusion decoder, transition-level Q-Former, pure-noise warm-up, simulation data, and two-stage training.
Physical language exhibits transferability and semantic structure, organizing videos by state transitions rather than appearance.
Limitations include lack of grounding in formal physical laws and coverage restricted to visually observable transitions.
02 From the paper
Abstract
We introduce PhiZero, a physical world model built around physical language, a compact discrete representation of world-state transitions. Existing physical world models typically predict future videos directly in pixel space, leaving the underlying world dynamics implicit within high-dimensional visual predictors. Motivated by humans' ability to abstract predictive structure from visual experience and organize it in natural language for explicit reasoning, we learn physical language from in-the-wild videos through self-supervision and use it to explicitly reason about how the physical world evolves. Accordingly, PhiZero adopts a reason-then-render paradigm: it first infers future world evolution as a physical-language sequence and then renders the inferred transitions into videos. Extensive experiments across generation and understanding benchmarks validate the ability of PhiZero to model physically coherent world evolution. We further show its potential for realistic and interactive world modeling, fine-grained action-conditioned simulation, and zero-shot motion transfer.