Gill Pratt Says Humanoid Robots’ Moment Is Finally Here

2026-04-02 · Source: IEEE Spectrum · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, medium

Summary

Gill Pratt, CEO of the Toyota Research Institute (TRI) and architect of the 2012 DARPA Robotics Challenge (DRC), discusses the current state and future of humanoid robotics. He notes that recent advancements are driven by AI breakthroughs, particularly in "system one" pattern matching, rather than mechanical improvements. Pratt highlights that current AI allows robots to learn tasks through demonstration, reducing the need for manual coding. He draws parallels between the data bottlenecks in robot learning and large language models (LLMs), advocating for "world models" and "system two" reasoning over mere pattern matching. TRI's work with "diffusion policy" and "large behavior models" represents significant "system one" advances, enabling robots to learn multiple tasks efficiently. Pratt also addresses the focus on legged humanoids, acknowledging their utility in human-designed environments but questioning their practicality in flat factory settings. He expresses concern about the "hype bubble" in humanoid robotics, fearing a "trough of disillusionment" due to overpromising capabilities that current "system one" AI cannot deliver without "system two" reasoning or human supervision.

Key takeaway

For Directors of AI/ML evaluating humanoid robot investments, recognize that current capabilities are largely "system one" pattern matching. Prioritize solutions that either integrate human supervisory control for "system two" decisions or focus on specific, well-defined tasks where pattern matching excels. Be wary of overpromising "reasoning" abilities, as this could lead to a "trough of disillusionment" similar to past autonomous driving hype cycles.

Key insights

Current robotics advancements stem from AI's "system one" pattern matching, but true reasoning ("system two") remains elusive.

Principles

• Robot utility now matches mechanical capability due to AI.
• Demonstration-based learning reduces coding for robot tasks.
• "System two" reasoning requires world models, not just pattern matching.

Method

Toyota Research Institute developed "diffusion policy" and "large behavior models" to train robots on multiple tasks, reducing data needs and improving performance through "system one" pattern matching.

In practice

• Explore "care-receiving robots" for elder care.
• Implement human supervisory control for "system two" decisions.
• Apply diffusion policy for robot behavior learning.

Topics

• Humanoid Robotics
• DARPA Robotics Challenge
• AI Revolution
• Robot Learning
• System One AI

Best for: AI Scientist, Robotics Engineer, Director of AI/ML

Related on AIssential

• Physical AI Matters More Than Humanoid Robots — Physical AI integrates perception, reasoning, and action to enable AI systems to interact effectively with the real world.
• Xiaomi EV’s Next Self-Driving AI Explained — Autonomous driving models can achieve superior performance by shifting from linguistic reasoning to a latent spatial-temporal mathematical representation.
• IEEE Transactions on Cognitive and Developmental Systems, Volume 18, Issue 1, February 2026 — The issue highlights advancements in embodied intelligence, human-robot interaction, and cognitive AI systems.
• This AI Watches Your Every Move — AI advancements are driving real-time coding, consumer wearables, and fully autonomous humanoid robotics, reshaping labor and production.
• HANDOFF: Humanoid Agentic Task-Space Whole-Body Control via Distilled Complementary Teachers — HANDOFF uses distilled multi-teacher control for robust, intuitive humanoid whole-body command, enabling diverse manipulation and locomotion.
• Proprioceptive-visual correspondence enables self-other distinction in humanoid robots — Proprioceptive-visual correspondence enables robots to distinguish self from others and learn a 3D self-model without prior knowledge.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.