Back to blog

What breaks when a robot meets the real world?

Thắng Đoàn
Thắng Đoàn

You funded a robot that reasons brilliantly, and it got lost in your own hallway. The fix costs months and money because the failure is not intelligence. It is the part nobody demos.

Easy things are hard

The tasks your roadmap treats as trivial are the expensive ones. In 1988, Hans Moravec observed a strange pattern, now called Moravec's paradox. Per Wikipedia, it is "comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility." The reason is evolutionary age. Skills like walking and recognizing faces took millions of years to refine, while abstract reasoning is evolutionarily recent. So the tasks your roadmap treats as trivial are the expensive ones.

  • Checkers and math: computationally cheap. Walking and seeing: hard.
  • Skills that evolved over millions of years are the ones machines handle worst, per Wikipedia.

The map that builds itself

You need a map to know where you are, and you need to know where you are to build the map. A concrete example is SLAM, simultaneous localization and mapping. Per Wikipedia, SLAM means a computer builds or updates a map of an unknown environment while simultaneously tracking its own location inside that map. It looks like a chicken-and-egg problem: you need a map to know where you are, and you need to know where you are to build the map. Wikipedia notes that approximate solution methods exist, and they work in tractable time for certain environments.

Why your demo hides the cost

A demo in a controlled space proves the easy half. Robotics, per Wikipedia, is the interdisciplinary study of the design, construction, operation, and use of robots. That word interdisciplinary is the trap. Reasoning software improves fast because it runs on commodity compute. Perception and mobility improve slowly because they cross fields: they must handle physical mess like light, surfaces, and unexpected objects, and no single discipline owns that problem. A demo means the robot runs in a controlled space you prepared. Deployment means it runs in spaces you did not. The demo proves the easy half. The expensive half only shows up on site.

  • Wikipedia notes DARPA's 2005 Grand Challenge winner Stanley ran SLAM as part of its driving system.
  • Budget for perception and mobility, not just the reasoning layer.

What this does not solve

Moravec's paradox explains where difficulty lives, not how to remove it. The supplied sources give no cost figures, no deployment failure rates, and no vendor comparisons. SLAM being solvable in approximate, tractable time for certain environments means it is not guaranteed for yours. You still have to test in the messy places where the robot will actually run.

Which parts of your product depend on perception skills you have only tested in clean environments?
If the hard half is physical rather than cognitive, is your hiring and budget split still right?

List every place your system must perceive the physical world, and mark which ones you have never tested outside a lab.

Sources

Wikipedia: Moravec's paradox

Wikipedia: Robotics

Wikipedia: Simultaneous localization and mapping

Share:

Recommended for you

Enjoyed this article?

Subscribe for new articles. No spam. Unsubscribe anytime.

By subscribing you agree to receive the newsletter. No spam, and you can unsubscribe anytime.