this just.. makes no sense. this doesn't play to Tesla's strengths or synergize with their existing R&D. Sensors for proprioceptive feedback are still immature, reinforcement learning for robotic motion is still in its infancy (Boston Dynamics did it the old fashioned way with differential equations and tons of work.) ...and you don't need any of that for cars, even with Autopilot. Turning steering wheels and accelerating is trivial compared to walking. Computer vision, sure, maybe, but that's a lot closer to solved than controlling a humanoid robot.
It plays exactly to Tesla’s strengths. It’s a completely unrelated futuristic area that can be used to raise a lot of funds and act as justification for an incredibly out of touch with reality evaluation of the company.
Also, all of the problems you mention are easily solved by putting Elon on a stage to tell potential investors that they’re already solved. Just like Tesla’s self driving system on the market now.
I think the humanoid aspect helps lay-people (and hedge fund managers) stay confused as to what the current scope of AI is. When people see humanoids, they think AGI, not DNNs.
The presentation was hilarious. He didn't even know what to say about it, just threw the slide up and a little dancing human in a robot outfit.
Just another distraction I think they threw into the agenda at the last minute. Won't be the first vaporware at these Tesla media days (has anything presented at these ever happened?)
So ... maybe they don't want to go for LIDAR (in their cars), because they wanted to build a humanoid robot anyway and felt that a LIDAR on a mobile platform's head is infeasible (?) - so they skipped the sensor fusion problem, and want to solve the harder problem first.
... and maybe go back to LIDAR later? Just guess work.
