AIToday
RoboticsRobotics & Automation NewsPublished: Nov 17, 2026, 01:00 JST

Moravec’s paradox: why physical AI needs touch, not just vision

Send on LINE
Moravec’s paradox: why physical AI needs touch, not just vision

3 Key Points

  1. What happened

    Ron Zukerman of Vishay Precision Group argues in a commentary that humanoid robots stall at the moment of contact, when cameras lose sight of the surface and only force sensing closes tolerances of tens of microns.

  2. Why it matters

    Without precise force feedback, Zukerman says machine-learning policies learn to ignore touch entirely and lean on vision, which limits tasks like connector insertion and threaded assembly in commercial sites.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Zukerman’s argument starts from evolutionary history: sensory processing, balance, and fine motor skills evolved over hundreds of millions of years and operate below conscious awareness. Abstract reasoning, by contrast, is a thin, crude layer that is easy to reverse-engineer into software. That helps explain why computers mastered chess and math before folding laundry, cracking an egg, or picking up a mug in an unfamiliar kitchen — tasks called some of the toughest problems in AI and robotics.

The piece then narrows to the point where visual predictions meet physical resistance. Camera-based vision provides spatial awareness, but contact cuts line of sight. Force and torque sensing turns a hard perception problem into a controlled physical search: the robot moves until it makes contact, probes until parts line up, and stops when resistance changes. Mass, friction, softness, stiffness, and grip proximity do not show up in pixels, so vision-trained models struggle to predict how objects push back.

On hardware, the piece warns that imprecise force feedback does more than degrade performance; it can cause the control model to delete touch as an operational modality. Preventing that requires unit-to-unit repeatability, long-term zero stability, kilohertz bandwidth, and deterministic latency, with a full multi-axis force-torque sensor distinguishing routine resistance from an impending jam. The broader point is that scaling physical systems demands different discipline than scaling software, where performance grows through model size and compute replication.

FAQ
What is Moravec’s paradox?
It is the inverted reality that tasks humans find intellectually demanding are computationally easy for machines, while physical interactions we perform without thinking are brutally hard.
Why can’t cameras handle manipulation alone?
Once a robotic gripper or hand closes around an object, cameras lose line of sight to the contact surface, so the system effectively goes blind precisely when physical interaction begins.
Robotics & Automation NewsRead Original Article
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleDraganfly to pursue drone manufacturing in the Philippines