Musk’s recent demo of FSD v12, where it tried to proceed through a red light, is a great example of how an “ordinary” situation can become dangerous when it’s outside of the training data. And even in the best case scenario, end-to-end systems won’t provide the safety auditability that you really want in a system that can be dangerous. May Mobility's system isn't limited by the training data collected ahead of time. Our cars generate training examples relevant to the vehicle’s current situation and learn from them on-the-fly. All of this happens on the car, in less than 200ms. We're the only AV company with this kind of real-time learning approach, and I think it's the key to building a scalable and safe system.
Are you imagining all the potential paths everyone in the scene can possibly do and plan the optimal path given the statistical probabilities?
I still remember the time we stopped at a stop sign and the car waited, because it correctly predicted that the vehicle on the cross road was going to run the stop sign!
Another reason why you're a world leader in AV development!
CEO and Co-Founder at May Mobility
1moat about 19'55 https://www.theverge.com/2023/8/28/23848882/elon-musk-tesla-fsd-v12-demo-red-light-zuckerberg-house