A Path Toward Physical Intelligence: Seeing AI's Full Potential

TLDR

• byFounders argues specialized cognitive robots, not humanoids, are the near-term opportunity as foundation models become accessible to any startup founder.

Key Takeaways

• Robotics has evolved through four waves; Wave 4 (cognitive robots using AI and vision for autonomous decisions) is now underway, enabled by accessible foundation models like π0, GR00T, and AutoRT.
• Specialized startups can outcompete humanoid players by solving narrow use cases well, collecting domain-specific training data, and avoiding the costly components humanoids require.
• Imitation learning via ALOHA and Mobile ALOHA can train bi-manual robots on end-to-end tasks with as few as 50 real-world demonstrations, with all code and hardware designs open-sourced.
• Over $100 billion was invested in robotics startups over the past decade; funding is projected to reach $12 billion in 2024, a roughly 40% increase from 2023.
• Priority verticals for cognitive robots include delivery, retail/groceries, cooking, and laundry, chosen because data and skills transfer across sectors and unit economics are under pressure.

Why It Matters

• Robotics foundation models from Google (AutoRT), Physical Intelligence (π0), Nvidia (GR00T), and Meta FAIR have moved from proprietary lab assets to tools any founder can build on.
• Unlike LLMs, robotics lacks large open training datasets, so specialized startups that collect domain-specific data while serving real customers hold a structural data advantage over generalist players.
• Demographic pressures, staff shortages, and thin margins in food service, retail, and logistics create immediate pull for cognitive automation that rule-based legacy robots cannot deliver.

byFounders · 2026-02-12 · Read the original