starVLA/starVLA

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

3530 stars · 466 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

StarVLA is an open-source, Lego-like modular codebase for developing Vision-Language-Action (VLA) models for generalist robots. It unifies model components, training recipes, benchmark evaluation, and deployment interfaces so researchers can plug-and-play modules and iterate rapidly.

Use cases

• train vision-language-action models for robot manipulation
• evaluate robot policies on LIBERO, SimplerEnv, RoboCasa, RoboTwin, and BEHAVIOR benchmarks
• swap model backbones like Qwen into a VLA training pipeline
• deploy trained policies to real robots via a unified WebSocket interface
• prototype new robot foundation model architectures with plug-and-play components

When to choose

• you are researching or developing VLA / robot foundation models and want modular, reproducible training and evaluation
• you need consistent benchmarking across multiple robot simulation environments
• you want to bridge simulation-trained policies to real-robot deployment without rewriting serving logic

When to avoid

• you need a production-ready, commercially licensed robot control stack (license is non-standard)
• you only need simple robot control without learning-based policies
• you require a stable, long-term-supported API rather than an actively evolving research codebase

Facets

framework · maturity active

machine-learning deep-learning robotics agent-framework llm-training robotics machine-learning artificial-intelligence simulation python vision-language-action vla embodied-ai robot-learning foundation-model modular-framework benchmark-evaluation sim2real gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="starVLA/starVLA")