TECH010: The Real Robotics Timeline w/ Ken Goldberg (Tech Podcast)

What It Covers

Ken Goldberg, leading robotics researcher, explains why humanoid robots face major technical barriers in manipulation and dexterity despite advances in mobility, challenging inflated timelines from companies promising household robots within five years.

Key Questions Answered

• •Robot Data Gap: Training language models required 100,000 years worth of readable text data, while robot manipulation data barely exists and must be generated from scratch through physical interaction, creating a massive development bottleneck that simulation cannot fully address.
• •Manipulation vs Mobility: Robots excel at navigation and walking with advanced quadrupeds and bipeds performing backflips, but picking up and manipulating deformable objects remains extremely difficult because humans use 15,000 sensors per hand to detect subtle forces, slip, and deformation that robots cannot replicate.
• •Simple Grippers Over Complex Hands: Commercial success comes from specialized simple grippers like suction cups that sorted 100 million packages at Ambi Robotics, not human-like hands with 22 degrees of freedom, because reliable task-specific solutions outperform general-purpose dexterity attempts for decades.
• •Visual Tactile Substitution: Surgeons perform complex manipulation like suturing without tactile feedback by using cameras to observe tissue deformation, suggesting robots might achieve dexterity through advanced vision systems rather than attempting to replicate human touch sensing, which remains technologically intractable.