OpenPI

Overview

OpenPI is an open-source library from Physical Intelligence that provides pretrained vision-language-action (VLA) models, specifically π₀ and π₀.5. These flow-based models are trained on large-scale robot data and support fine-tuning for custom robotics tasks.

Best for

Best for
Robotics researchers and engineers needing pretrained VLA models for manipulation tasks

Use cases

• Fine-tuning pretrained VLA models for specific robot manipulation tasks
• Building vision-based robotic control systems with natural language instructions
• Researching and experimenting with flow-based action prediction in robotics

Notes

OpenPI is an open-source library from Physical Intelligence that provides pretrained vision-language-action (VLA) models, specifically π₀ and π₀.5. These flow-based models are trained on large-scale robot data and support fine-tuning for custom robotics tasks.

12,128 stars on GitHub. Last updated 2026-05-05. Licensed Apache-2.0.

Use cases

• Fine-tuning pretrained VLA models for specific robot manipulation tasks
• Building vision-based robotic control systems with natural language instructions
• Researching and experimenting with flow-based action prediction in robotics

Pros

• Pretrained on large-scale robot data, reducing need for extensive custom data collection
• Open-source with permissive license and active community (12k+ stars)
• Supports fine-tuning, enabling adaptation to new tasks and environments

Cons

• Requires significant computational resources for training and inference
• Limited to robotics domain; not applicable to general-purpose vision-language tasks
• Documentation and examples may be sparse for beginners

Indexed from awesome-llmops and enriched against its public facts.