How to close the 100,000 year data gap in robotics? (1/3)
I attended
@Ken_Goldberg's talk yesterday without knowing much about robotics, and I found it super interesting!
Apparently, the bitter lesson follows you everywhere...
Data has solved language and vision. Does it also solve robotics?
Yes, it will but with some caveats.
Building effective robotic systems comes down to two things:
1️⃣ Data (there is so little in comparison to what's used to train LLMs)
2️⃣ Good old fashioned engineering (GOFE) - modularity, algorithms, and metrics → just designing good systems
We've seen GOFE in AI systems shifting to modular systems (like DSPy!) and biological control systems like the brain.