physical-superintelligence-lab/Psi0

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

2802 stars · 91 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Psi-Zero (Ψ₀) is an open vision-language-action (VLA) foundation model for dexterous humanoid loco-manipulation, combining a Qwen3-VL backbone with a flow-based diffusion transformer action expert and an RL tracking controller. The repository provides model weights, training data, and code released alongside an RSS 2026 paper.

Use cases

• control a humanoid robot for dexterous loco-manipulation tasks
• fine-tune a VLA model on a small set of teleoperated robot trajectories
• pre-train a vision-language backbone on egocentric human videos
• benchmark humanoid manipulation policies against baselines like DreamZero or SONIC
• run whole-body action chunk prediction on a humanoid embodiment
• research data recipes for scaling humanoid robot learning

When to choose

• you need an open state-of-the-art VLA model for humanoid whole-body control
• you want to fine-tune a foundation model with as few as ~80 trajectories
• you are doing research on humanoid loco-manipulation or egocentric-video pre-training
• you need a diffusion-transformer action expert architecture to build on

When to avoid

• you need a production-ready industrial robot control stack
• you lack GPU hardware or a humanoid robot platform for deployment
• you need a general-purpose home assistant robot out of the box
• you require a permissively documented, fully turnkey training pipeline with extensive support

Facets

library · maturity active

machine-learning deep-learning llm-training robotics transformers rag robotics machine-learning artificial-intelligence deep-learning large-language-models python vision-language-action humanoid-robot loco-manipulation foundation-model diffusion-transformer egocentric-video vla teleoperation fine-tuning research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="physical-superintelligence-lab/Psi0")