The obstacle avoidance sensors on my quadcopter wouldn’t have let me hit that pole. They are just optical cameras, and work at speeds of ~22mph and below.
It seems nuts to me that Tesla can’t do better than a ~$1,500 quadcopter.
It seems nuts to me that Tesla can’t do better than a ~$1,500 quadcopter.
That's probably because you've mischaracterized the issue. It's a streetlight. Obviously Teslas see streetlights. Go for a ride in a friends car (not even one with FSD Beta) and look at the screen. You'll see streetlights rendered by the hundreds. It's basically perfect. It failed here, not because it couldn't see "a pole" but because of a much more specific bug. Specific bugs can, almost always, be fixed.
Here's the latest, even when it finally succeeds, it is not very convincing or consistent.
Or do you think that the progress we're seeing currently will just stop? You think that video is as well as the car will ever handle that particular situation to your satisfaction and no further progress is possible?
Most software has to deal with a limited input domain, and when mistakes are made, they are tolerable. And even in the case of software in limited domains, critical bugs remain in commercial software that have been around for decades, with new bugs cropping up as the software is maintained.
Now look at FSD: What is the input domain? The entire real world and the incredible diversity of variety of it. And what is the repercussions of a bug? Death or injury.
So like I said, these FSD systems should be designed with a kind of "kernel" vs "user" space paradigm. The "kernel" has a mission to absolutely prevent "user" space from crashing the system. In my view, that would mean a backup sensor suite and software design whose highest priority is to avoid collision, rather than driving or navigating. All of Tesla's ML systems should be in the "user" space. They view the world, make predictions to navigate, but any control inputs they yield should always be overridden by the "kernel" if a collision is pending.
The vision has to be good enough to handle this type of situations with acceptable error rate.
Also, sensor fusion is difficult - LIDAR does not see colors, Radar is just weird.
Adding more complexity will eliminate some bugs, and result in new bugs.
That's non-falsifiable. You can end a discussion that way, but not win an argument. Let's just leave it at "FSD beta appears to be improving rapidly" (c.f. the public history of this stuff since the first public beta users received the build in March) as my response.
I don’t think I have.
The OA feature on my quadcopter positively prevents you from flying into obstacles, even intentionally. It’s not perfect - power lines, wires, and bare branches aren’t seen 100% of the time - but I’ve never had it trigger a false positive.
Why wouldn’t there be a similar system at play here, limiting the actions of the FSD?