Two weeks ago we were witness of the Autonomous day by Tesla. It was kind of futuristic introduction of the new time. 4 hours long deeply technical presentation dove deep into the future of the way we commute. Here are 13 things, that I think worth mentioning for the self driving cars:
- In the beginning of 2016 there was no chip dedicated exclusively to self driving
- It took Tesla 3 years to design and develop FSD (full self driving) from ground up and ramp up to full production and integration with Model S, X and 3
- The main goals were to stay below 100W for the energy consumption and at least 50 TOPS of neural network performance
- Full redundancy – there are two chips which are constantly communicating and verifying the decisions
- FSD is 4 years ahead of the competition in terms of self driving capability.
- It (the new chip) is capable of 144 Trillion operation in one second for the 2 separate neural networks and consumes 72W of power
- This can handle video with 2300 FPS (frames per second) in comparison with the older nVidia chip used with only 110FPS
- To make neural networks work we need 3 essential things:
- Large dataset
- Varied dataset
- Real dataset
- Train, deploy, boost, test, label, repeat
- The fleet labels data automatically
- The system runs in so called shadow-mode and after a series of iterations it finally goes live.
- It is all about long tail : 99,9999…%
- Until the end of 2020 (ET – “Elon time”) Tesla will have a million autonomous taxis on the road and will disrupt the hole idea of transportation
