EMBODIED AI
Why Robots Struggle With Toddler Tasks?
A model can solve complex problems in language, yet fumble picking up a cup. To us, the cup is effortless ;)
Why a Toddler Beats a Robot
A chess engine can crush a grandmaster, yet the same AI struggles to pick up a mug. We find it effortless, and that's the catch.
Moravec's Paradox
Hans Moravec noticed that abstract reasoning is often easier for computers than perception and movement. We forget how hard the physical world is because we've spent our lives learning it. A robot starts from zero.
"Put the Mug on the Shelf"
Simple for us. For a robot, it means finding the mug, judging its position, planning a path, choosing a grip, and fixing mistakes on the way. One small perception error can end in a dropped mug.
The Real Challenge
Making a model smarter may not be the hard part. Making it reliable when things go wrong is. A robot that works in a demo is impressive. One that recovers from failure is useful.
That's what draws me to embodied AI: systems that can see, think, act, and adapt in the real world.
