openpi
A frontier robot policy, weights and all
What it is
Physical Intelligence's open release of its flow-based vision-language-action models. π0 is pretrained on 10,000+ hours of robot data; π0.5 adds open-world generalization via 'knowledge insulation'; π0-FAST is an autoregressive variant. Base models ship with DROID, ALOHA, and LIBERO checkpoints.
Why it's interesting
A rare fully open (Apache-2.0, weights included) release of a frontier generalist robot policy from a leading robotics-foundation-model company — the kind of model that's usually locked away entirely.
Use cases
- Generalist manipulation policies
- Fine-tuning a real VLA on new robots and tasks
- Benchmarking on standard robot-learning suites
Who it's for
Robotics researchers, advanced ML engineers, robot-learning labs
Setup
Advanced. A >8GB GPU for inference; a 4090 for LoRA fine-tuning; A100/H100-class for full fine-tuning
Limitations & cautions
Developed for PI's own robots, so it may need adaptation on standard platforms, and compute needs are heavy.
Editorial takeaway
When a robotics-foundation-model company open-sources its actual policy, the field moves. This is that release.