ZeApelido
Active Member
Nobody supports the idea that humanoid robots are around the corner. NOBODY. I am not arguing on the NN side of things. I am sure on the synthesis of mechanical & NN side of things. The thing that makes a human so neat just from a science pov. So lets look at it from an investor POV. Could you put any value today?
The goal of the society that promotes humanoid robotics is to be there by 2040 and to have robots play world class soccer by 2050. Literally not a single researcher on the planet thinks it will be there by 202X. None. It's not that I had to do much searching to confirm that since I last looked at it all things have continued at the same pace they had been, which is good solid progress, year by year with thousands of researchers across the world contributing. Korea to Japan to Germany to USA. I do not see Tesla having any special competency and as an investor Eyes Wide Open. A lot of people have been working on this for decades. The fundamental science and research is not held by tesla, it was not discovered by tesla and Tesla won't own any specific competency there.
So, my phd was in dynamic modeling of humanoids (more for trying to mimic humans than making robots) so I have some experience and opinions for why Tesla is well suited here:
1. The historical fundamental research will not matter much.
The history is based on developing the "equations of motion" that prescribe the state of a robot which is very complicated math. Basically a whole lot of equations / matrix math that tells you what joint torques to apply to keep the robot balanced given any state of positioning and velocity of movement. Except they are very hard to get working well because the real world is different from the simulated world and it is very hard to model correct the interaction forces with the ground and other objects.
Enter neural nets. Much like classic computer vision techniques fell aside to deep learning CV, much of the classical control of robots will give in to deep learning. And all those ressearchers will not be happy about that.
There are two many variabilities and complexities in actually robot dynamics vs what classical modeling can do, and way toward superior robust performance of control logic is going to be through statistical learning.
Guess which companies are going to have the resources for provide the best chance for advancement in statistical learning for robots? The big tech companies, Tesla, Facebook, Google.
2. Hardware advantage
The cost advantage Tesla will create is obvious vs any robotics startup or any other tech company. What about sensors and performance? Well batteries and computer brain - advantage Tesla. Engineering to minimize latency between limb sensors and 'brain' - advantage Tesla. Does someone have some novel pressure / tactile sensor that is superior and game changer? Tesla just buys them.
3. Data
Kind of similar to FSD though not as much, superior robotic performance will happen when there is a fleet of child robots are deployed in some capacity, fail, and have that data transmitted back to a server for enhance learning. Update firmware until they become teenagers, etc... This sort of robotic learning at a fleet level isn't done yet, because no one had the capabiilty to do all the pieces. Ahem.
The hubris of the robotics community will not stop Tesla from succeeding. They have all the pieces.
Instead of asking the inverse, I would ask "Where does Tesla not have an advantage in modern day robotics"?