Robot maker · registered

Physical Intelligence

Corporate marketmanufacturinglogisticsCurated record

pi.website

Physical Intelligence builds general-purpose Vision-Language-Action (VLA) foundation models (π-series) that give any robot body the ability to perform open-world manipulation tasks without task-specific re-training. The π0 architecture (Oct 2024) combines a pre-trained vision-language backbone with a flow-matching diffusion action head, enabling dexterous multi-step behaviours; laundry folding, vegetable peeling, box assembly; across heterogeneous embodiments. π0.6 (Nov 2025) introduced RECAP, the first self-improvement loop allowing a deployed general-purpose model to learn from its own field rollouts. π0.7 (Apr 2026) added multi-scale embodied memory and online reinforcement learning; as of mid-2026 no paying commercial customers have been publicly announced.