> Software has been holding back robotics for a while.
Honest question; have you ever spent more than 3 months working in the field of modern robotics?
Contextually I find this statement pretty funny. I've worked with and witnessed people moving mountains of software to make robots work. The primary constraint of robotics is and always will be physics.
Nope haven't worked on robotics (at all) so you may be right.
Can you provide more insight? We see things like the Boston dynamics robots that look cool but never seem to materialize in an actual product. My assumption is that while they can get those robots to do cool things, its hard to generalize and make it repeatable in an unknown environment.
Here is where I think Elon is trying to add NN so that the data from robots can be used to train and make it repeatable. This is why he considers Teslas experience with self driving to be a start.
Software has been holding back robotics for a while.
Tesla pouring money training humanoids with state of the art neural nets could lead to something disruptive.