I personally love the progress they are making on some of the key hardware that actually makes it practical - increasing DoF of the fingers, force-torque sensors, compact tactile sensors in the fingers, etc.
Given good hardware, planning and control to move around is relatively easy in 2023, at least for a company like Tesla with a ton of resources. They need high-fidelity simulators, lots of GPUs, lots of human demonstration data, and expertise to train robots end-to-end - they have all of them, including some super smart researchers who spent their grad years doing exact same stuff.
The finger actuators seem to be in the hand instead of connected to the middle of the arm. Things like that make me think they’re not really thinking this through, or borrowing from actual anatomy. Their approach on the AI also seems to be demo oriented, meaning they’re just training little models to do something simple that they can show in the video and that’s it
I really hope Tesla is going to sell this as a commodity robot even though their AI software isn't perfect.
Or even the motors and actuators separately.
If the video is real, what they've done on the hardware side is much more remarkable.
Just the potential for teleoperation is worth a few billion dollars. Would buy even if it has 0 built-in AI capability but I could run my own software on it.
They shouldn't wait for it to be perfect. Robotics will move ahead a lot more faster with good hardware.
4 comments
[ 0.20 ms ] story [ 23.2 ms ] threadGiven good hardware, planning and control to move around is relatively easy in 2023, at least for a company like Tesla with a ton of resources. They need high-fidelity simulators, lots of GPUs, lots of human demonstration data, and expertise to train robots end-to-end - they have all of them, including some super smart researchers who spent their grad years doing exact same stuff.
Or even the motors and actuators separately.
If the video is real, what they've done on the hardware side is much more remarkable.
Just the potential for teleoperation is worth a few billion dollars. Would buy even if it has 0 built-in AI capability but I could run my own software on it.
They shouldn't wait for it to be perfect. Robotics will move ahead a lot more faster with good hardware.