Soft Robotics Podcast What AI Robotics Demos Don’t Show: The Missing Piece in AI Robotics
Apr 1, 2026
Phil Trubey, an AI and robotics industry expert with decades tracking the field, discusses why humanoid demos hide how robots actually learn. He compares fast motor skills to slower planning, explains why fine dexterity like tying shoelaces is hard, and explores teleoperation, simulation, training bottlenecks, neuromorphic chips, and the gap between flashy hardware and practical deployed robots.
AI Snips
Chapters
Transcript
Episode notes
Why Robot Demos Hide The Learning
- Robot learning is rarely shown because the training process is cumbersome and unglamorous.
- Phil Trubey explains companies hide slow teleoperation and simulation training that produce overfit demos rather than continuous learning footage.
Tying Shoelaces Shows The Motor Learning Gap
- Phil uses tying shoelaces to show system one motor skills are subconscious and hard to verbalize.
- He demonstrates fingers adapt via cerebellum-level control, highlighting why tactile dexterity is tough for robots.
Use Teleoperation Then Simulation For Training
- Train fine manipulation via teleoperation and supplement with physics simulation for gross motor RL.
- Phil advises tele-op humans control fingers to collect data, then use sim RL for walking and other large-scale skills.