Physical Intelligence vs Adept AI

About

Physical Intelligence (pi) is a San Francisco-based robotics AI company founded in 2024 to build foundation models for general-purpose robot control. The company was founded by a team of leading robotics and AI researchers including Sergey Levine, Chelsea Finn, Karol Hausman, and Brian Ichter, who previously held senior positions at Google DeepMind and leading academic robotics labs. Physical Intelligence's mission is to develop a single AI model capable of controlling any robot hardware to perform any physical task — the robotics analog of what GPT-4 did for language understanding.\n\npi's core product is a general-purpose robot foundation model trained on diverse robot demonstrations across tasks, hardware types, and environments. The model supports Multi-Scale Embodied Memory (MSEM), an architecture designed to give robots persistent task memory across long-horizon manipulation sequences. Rather than training separate models for each robot type or task, pi's approach generalizes across hardware configurations, enabling faster deployment across the heterogeneous robot ecosystem that exists in warehouses, manufacturing facilities, and research labs.\n\nPhysical Intelligence achieved a $5.6B valuation shortly after its founding and is reportedly raising an additional round at a valuation exceeding $11B — a remarkable trajectory for a company less than two years old. The company has received backing from prominent investors including Sequoia and Jeff Bezos. pi represents the bet that physical world AI — controlling robots in the real world — is the next frontier after language and image models, and that a foundation model approach will ultimately outperform the task-specific robot programming paradigm that has dominated the industry.