Understood.
I think that understates what companies can get from Nvidia. I believe that Nvidia DRIVE Hyperion includes both the hardware platform and software, including ML models. But Nvidia is only pitching it as "level 2+" at this point. Even Waymo, the clear leader in autonomous driving, readily acknowledges that they have quite a bit of R&D that still needs to be done. For companies not wanting to invest a few billion trying to build their own autonomous driving, Nvidia offers the most obvious technology to license. But until autonomous driving is a solved problem, it's far from clear who the viable technology partners will end up being. Waymo seems like a clear bet. Nvidia and Tesla both have a willingness (and financial capacity) to dump endless billions into the tech as well, and so will probably someday succeed. But as fundamental research in AI continues to improve the base capability of the technology, the barrier to building an autonomous driving system will slowly drop. Probably the smart thing to do for most companies is to avoid dumping in huge somes of money at this moment when the state of the technology isn't quite ready to offer the capability.
Lidar and radar are complementary to cameras. Each has unique strengths in a range of conditions. Lidar and radar can see in the dark, radar can see in heavy fog, etc. I'll take all three, as I place a high value on our safety. I'm not interested cheapest solution.
I am not talking about the cheapest solution. I am looking for the most effective mix. When cameras, radar and lidar don’t all agree - which does the car choose? That’s an issue.
Yes, it's part of the problem to be solved.
It's true that combining data from multiple sensors poses a challenge. But that's not unique to radar or lidar. Two optical cameras with overlapping fields of view also need to solve the problem. The human brain has to do it with our eyes as well. It's actually an incredibly challenging problem, and even our highly evolved brains get it wrong on a regular basis. In terms of looking for the most effective mix, I believe that it's better to have multiple kinds of sensors, where one kind can be effective in types of sensing that another kind is weak at. More sensors makes the vehicle more expensive, but I personally am willing to pay more for the safety it can provide. I know that's easy for me to say, a guy who's planning to buy a Gravity for more than double what the average new vehicle costs. But even lidar has already come down enormously in price, and radar is already pretty inexpensive.
The reason Musk did not use LIDAR and got rid of Radar is exactly this reason- the computer power needed to figure out which is right or wrong means it will take him another 2 decades to get this right- and of course the expense.
Getting rid of one or two types of sensors that may disagree is not the solution. Better sensor fusion is the solution. Getting rid of sensors is a safety risk to save cost.
This is correct.
I think if Tesla didn't have their somewhat ambiguous marketing that tends to oversell FSD as being autonomous driving, then you might not find so many people pushing back against it. As an ADAS system, it's pretty good. There are some concerning incidents of it misbehaving too rapidly and severely for the driver to seize control. But to be fair, those incidents seem to be very rare. As an engineer, I prefer my life-critical systems to have more of a safety-first engineering approach than Tesla has taken. But I know not everyone sees it the way I do.
Curious how often you think FSD would crash without a human monitoring it.
I am not the average driver, and I don’t want to revert to the mean. Most people suck at driving, and roughly half are worse than average.
It took Rolls Royce many years to fail at building engines that were powerful enough to break the sound barrier at altitudes high enough that there was no sonic boom on the ground. In fact, they never did it.
I don’t need 100% safety. I just want it not to make mistakes I wouldn’t have made. I don’t stop randomly in the middle of the road by slamming on my brakes with nothing in front of me.
While I agree, and have the latest version of FSD (Supervised) in my Model X Plaid, my fear is rarely FSD anymore (rarely) -- it's the behavior of other drivers while FSD is engaged that frustrates me. THEY may cause a crash due to THEIR actions.
