Pi API A hosted robotics foundation-model API from Physical Intelligence. Teams upload their own robot data, fine-tune Pi models, and run cloud inference on their hardware so they can specialize a shared robot brain instead of training one from scratch.
Visit productFeatured media
Loading media…
View on XCosigns Received
“Congratulations to the PI team on the beta release of the PI API! Exciting to see this becoming accessible to more robotics teams. I remember the early days of working with @QuanVng and the PI team, when we had to constantly coordinate on data ingestion, and which dataset to use for each training run. Today, that entire process is automated. With the right config file, we can kick off a new training run in under a minute. At @DeftRobo, our focus is on collecting high quality, task specific data from real-world deployments, identifying failure modes, and rapidly feeding that data back into training. PI focuses on advancing foundation models. By working together, we can significantly shorten the cycle from data collection to production deployment without spending days engineering task-specific algorithms for every new task. I'm increasingly bullish on this partnership between model labs and deployment companies. Scaling robotics takes both. Excited to see more teams leverage the PI API to bring robotics into the real world across many more applications!”
“Welcome to the dawn of physical intelligence. One foundation model, any (eventually) embodiment, fine-tune and deploy. This is where true pretraining and scaling for robotics begin.”
“Even a year ago I don't think I would have believed that PI models can run on such a huge range of different robots.”
“Working with @physical_int, we are bringing general purpose robotics models into the places that need them most —- the factories that make things we all need and use. 150 factories using Formic robots and counting!”