Remus
Active Member
It seems Tesla already has some machine learning in its manufacturing processes:
From an employee on Linkedin - Senior Machine Learning and Controls Engineer at Tesla
I presume Tesla ultimately aims to move much of the manufacturing code to machine learning. OpenAI made a breakthrough here last year with a new method to train a robot in simulation that was much better at transferring to the real world than prior methods. Their idea was that rather than trying to model the physics of the real world accurately (which makes your simulation very brittle to real world imperfections and complexities not captured by your model), they instead varied the laws of Physics in their simulation and trained a robot that could perform well in any of these simulated conditions. This AI robot was much more robust at dealing with the variations between the real world and the simulation. Still a long way to go though. Learning Dexterity
I'm quoting this for the discussion about the cost for building the car. My point is multiple sources indicated that Tesla is not standing still on the manufacturing front. Their efforts will bear fruits that reduce the cost.
Elon saying "humans are underrated" but they are still trying to reduce human labor on the line. Computer vision, for which they have tremendous expertise, is widely used. Tesla is already using computer vision on quality control. That's the low hanging fruits many AI companies are also working on, just take a look at Andre Ng's landing.ai.
There is also information indicating that they are working on using vision to guide the robot arm. The eye hand coordination part is harder than vision based quality control, but it's much simpler than self driving.
I am optimistic that they can continue to lower the cost as their technology progress