Physical Intelligence

Overview

Physical Intelligence (π) is a robotics AI startup founded in 2024, developing general-purpose robot foundation models. Founded by a team from Google DeepMind, UC Berkeley, and Stanford, the company raised the largest seed funding round in robotics history.

Item	Details
Headquarters	San Francisco, CA
Founded	March 2024
CEO	Karol Hausman (formerly Google DeepMind)
Funding	$1.07B (Seed: $70M, Series A: $400M, Series B: $600M)
Valuation	$5.6B (as of November 2025)

Founding Team

Co-Founders (7 members)

Name	Previous Role	Position
Karol Hausman	Google DeepMind (RT-2), Stanford Adjunct Professor	CEO
Sergey Levine	UC Berkeley Professor (RL Expert)	Chief Scientist
Chelsea Finn	Stanford Professor (MAML, Meta-learning)	Research
Brian Ichter	Google DeepMind (RT-2)	Research
Lachy Groom	Stripe (Former Head of Stripe Issuing)	Business/Product
Adnan Esmail	Anduril (Former Head of Electrical Systems), MIT	Engineering
Quan Vuong	Robotics/RL Researcher	Research

Key Investors

Seed/Series A:

Thrive Capital, Lux Capital
Khosla Ventures, OpenAI
Jeff Bezos, Sequoia, Bond

Series B:

CapitalG (Lead), Lux Capital
Redpoint Ventures, Sequoia Capital
T. Rowe Price, NVIDIA (NVentures)

pi0 Model

Key Features

Item	Details
Parameters	3B (3 billion)
Architecture	PaliGemma + Flow Matching
Core Technology	Action Expert with Flow Matching
Open Source	Released February 4, 2025 (openpi)

Flow Matching Approach

Uses Flow Matching instead of Diffusion:

[Noise] ──Flow Matching──→ [Action Chunk]
        (faster inference)

Faster inference than Diffusion
Well-suited for continuous action space
Capable of learning multi-modal action distributions

Performance

Trained on 68 tasks across 7 robot embodiments
Over 10,000 hours of real-world robot data
Zero-shot generalization capability
Superior performance compared to single-robot policies

Data Collection

Diverse Robot Platforms

Single-arm robots
Bimanual arms
Humanoid upper body
Mobile manipulators

Data Characteristics

Cross-embodiment data
Diverse environments (homes, warehouses, offices)
Includes dexterous manipulation

Approach

”Bring GPT to Robotics”

LLM Success = Large-scale Data + Transformer + Scaling

Physical Intelligence Goal = Apply same approach to robotics

Three Core Principles

Scaling: More data, larger models
Generality: Not limited to specific robots/tasks
Foundation Model: Fast adaptation through fine-tuning

Roadmap

Timeline	Milestone
2024.03	Company founded, $70M seed funding
2024.10	pi0 release
2024.11	$400M Series A ($2.4B valuation)
2025.02	pi0 open source release (openpi)
2025	pi0.5 release (open-world generalization)
2025.11	pi0.6 release, $600M Series B ($5.6B valuation)
2025+	Commercial deployment