[gzer0]

Expense was part of the equation initially, however, through economies of scale, we eventually would have been able to reach a feasible price point. Cost has nothing to do with why Tesla is pursuing an optical-only system.

To get rid of the dependency on the radar sensor for autopilot, we generated over 10 billion labels across two and a half million clips. To do this we had to scale our offline neural networks and our simulation engine across 1000s of GPUs, and just a little bit shy of 20,000 CPU cores. We also included over 2000 actual autopilot full self driving computers in the loop with our simulation engine. And that's the smallest compute cluster.

[meheleventyone]

Those are very large numbers for something that doesn’t work very well.

[atleta]

So what's the point then? You said it's not expenses and then you explain how you think it caused you extra trouble/work/development effort. But what's the reason?

[stefan_]

That must be why complaints about phantom braking have gone through the roof since the switch away from radar.