Physical Intelligence develops foundation models and learning algorithms for robotic manipulation, with products including π0.7, a steerable robotic foundation model released in April 2026 that demonstrates compositional generalization—the ability to perform tasks robots were never explicitly trained on. The company's technology stack includes vision-language-action (VLA) models enhanced with multi-scale embodied memory for tasks exceeding ten minutes, reinforcement learning tokens for efficient online learning, and real-time action chunking systems. Co-founders include Sergey Levine, a UC Berkeley professor focused on AI for robotics. Physical Intelligence is based in San Francisco and has operated as a two-year-old startup as of April 2026. The company uses the WEKA Data Platform for AI model training and has demonstrated 10–15% improvements in training speed. Specific funding round details, additional named executives.