What I have realized is that most of robotics figure on twitter have never worked with robots. Everything he said is not controversial to a roboticist. Try unlocking your door with very thick gloves. That's the sensing robots are missing. Vision was never enough.
I've lost track of the reply chain, posting here: 1. We can’t claim "vision is all you need" until Physical Intelligence or other startups show 99.99% open-world success rates, not just cherry-picked demos. 2. Autonomous driving is a fundamentally simpler physics problem than manipulation. FSD's entire goal is avoiding contact. Manipulation is entirely about making contact, controlling forces, and managing micro-nuances where a millimeter of error equals failure. 3. As a daily FSD user, vision-only clearly struggles with edge cases like tight parking, where old-school ultrasonic or radar sensors easily beat it. If pure vision struggles to judge a static wall 3 inches away, relying on it to modulate precise forces for complex manipulation is a massive gamble.




