Physical Intelligence Announces Robot Training Method Recap and VLA Model π*0.6

The company developed π*0.6 using Recap, a method that trains robots in three stages: demonstrations, coaching and reinforcement learning. Physical Intelligence said π*0.6 achieved more than double the throughput and less than half the failure rate on tasks including making espresso, assembling cardboard boxes and folding laundry.

According to Physical Intelligence, Recap is a training method designed to address the cumulative errors that can arise from human demonstrations alone by using corrective interventions and autonomous experience. It targets tasks such as making espresso, assembling boxes and folding laundry, where robots need to operate quickly and reliably over continuous runs in the real world. The company says π*0.6 improved throughput and reduced the failure rate.