I have some experience here so I'm just going to chime in because it's fairly obvious that this post is really big on hype and fairly low on actual comprehension of what's behind this technology. I'm not a Phd student studying the latest in ML/AI but I have experience applying it to less complex problems so I can speak with some authority.
I'm not a skeptic, I just hate bs claims and Tesla right now is the king of bs claims. It's then perpetuated by people such as yourself with little more than a superficial understanding of what's happening and the hype train keeps moving.
I'll address your post piece by piece.
Quote:
Originally Posted by kring
Tesla’s goal from the beginning (part of the master plan) was to become 10x safer than a human.. they have pretty much achieved that. They did have some other lofty goals like robotaxi and full autonomy and they will continue to work toward that - but again, 10x safer than a human - not PERFECT.
|
They're not the leader in autonomous driving, Waymo is. Waymo is actually running robotaxis and only in areas of the country that have exceptionally easy weather to handle and roads that aren't as complex. Tesla has no autonomous service on public roads thus it's really hard to say they're the leader.
Quote:
Originally Posted by kring
Hardware wise, a tesla has everything it needs to be 10x safer than a human at driving, and can probably go well beyond that with further software improvements which they are doing as the mileage racks up from the ever growing fleet. A lot of people like to claim you need radar, sonar, lidar, 64 cameras.. but nope, they have all that is required for every possible legal driving scenario. Radar is inferior to cameras but inexperienced people keep thinking it’s needed because it sounds cool and others have it. Tesla ditched it because it added complexity and had zero benefits. They did say that if someone made a high-res and long distance radar they would implement it in future models, but that tech doesn’t exist yet. For some reason, the majority think that self-driving means “never ever ever makes a mistake or gets into an accident” that’s not at all accurate or reasonable to assume…. It will get in accidents, just not as often as humans.
|
This is really where the BS hits ridiculous levels. How can you possibly make such a claim as a layman? Where is the data to back this up? The companies that actually run Robotaxis use lidar, radar and vision. Human eyes have lots of advantages that cameras don't, higher resolution, better ability to see in extreme lighting conditions, and the ability to shift one's perspective by moving slightly or even something as simple as shielding your eyes from blinding sun. Cameras can't do most of that so you use other sensors to augment the gaps.
Telsa ditched radar because they couldn't source the components during the pandemic and told their customers it was a good thing. Now they're bringing it back and I can't wait to hear the BS story they feed their customers that they just told radar was useless 2 years ago.
https://electrek.co/2022/06/08/tesla...adar-confused/
Quote:
Originally Posted by kring
From a development & AI training perspective, you need a base. there is more value gained from making a camera-only based system work and investing in maximizing existing hardware.. then down the road adding more tech would be major improvements for little effortÂ… itÂ’s the base that you need to build out first and that has to be camera. So Tesla is taking the right approach, but like most critics & skeptics - they will not believe it until well after they have seen it.
|
You use a Camera based AI because sensor fusion is hard. It's not "easier" and "little effort" to go from a camera based system to one that incorporates data from multiple sensors.
Also, it's unclear whether or not their existing training data will help them or not going forward. As sensors change, it's not obvious to me or anyone outside the industry whether or not you can re-use existing training data at all.
The bottom line is that companies like BMW and Mercedes are going to take financial liability if their systems screw up and I suspect Tesla never will. I know which system I'd trust more.