Tesla FSD Beta 8 tested for self-driving performance in snowy conditions

- You weren't using the actual FSD software, but the live production autopilot, so it's Apples to Oranges.

- It's not clear that the transition from low quality to high quality full self driving will be linear. Most claims around anecdotes come with an unspoken assumption we ought to see steady improvement over time vs a step-function like behavior. The latter seems somewhat possible given the nature of how improvements in AI systems seem to happen.

- Beyond all this, without understanding the underlying cause of the behavior you mention, it's hard to say from a risk perspective if the car made the right choice. Unless you have a clear understanding of the data and mechanisms that led to the speed reduction, it's hard to be sure it's a bug beyond not doing such transitions in a way that are comfortable for passengers. (My understanding is adding smoothing to the control algorithm is somewhat separate from overall decisionmaking, and hence poor control algorithms may mislead you into thinking it's more dangerous.)

You might as well be comparing your experience driving a BMW or Cadillac with SuperCruise. Your experience on the old tech stack is irrelevant.

As it relates to the features, I agree that they aren't ready for what most drivers expect, either because they aren't reliable enough (beta) it because they aren't confident enough (TACC). With that said, FSD is a feature set, not a level of reliability. As a whole it eventually needs to reach a level of reliability, but there's no specific time horizon.

For Tesla, a feature moving from beta to alpha is what indicates reliability. In addition, that reliability may not come with the degree of confidence most people expect from a driver. Safe FSD could just end up as the AI equivalent to a student driver. It'll get you someplace safely, but there may be phantom braking, excessive waiting, and so on. If that's the case it would be nice if we could enable a beta version with more "normal" driving that we would have to keep an eye on.

The long term, macroscopic decision making involved in driving computers are nowhere near ready to drive better than I can in winter (or even in perfect conditions). I love my model 3, but the autopilot is simply not a good driver. It would fail any minimally rigorous driving test because it makes rookie mistakes and only drives in the exact moment. It cannot plan ahead more than even a handful of seconds. It stays in other driver's blindspots when there is plenty of room to adjust the relative positions of the cars. It accelerates into gaps that are about to not exist because other drivers' small hints scream they are about to change lanes, even if they haven't signalled yet. It gets confused about speed signs, often reading the off-ramp maximum as the speed and slams on the brakes. It plants itself in the dead centre of the lane even when other drivers encroach and then when it decides the situation has become dangerous it swerves violently and applies applies significant braking. Experienced, safe drivers give ground within the lane, preventing the dangerous situation from ever developing, and if the other driver does fully enter their lane are ready to calmly move over without creating an entirely different collision.

I'd feel a lot safer with that kind of data added to the current sensors.

Tesla's have 5 inches clearance tops. FSD doesn't need to be able to drive through two foot drifts.

It depends. If they're not recording the right things, they have the same 0 bytes of data as everyone else.

There's a not insignificant chance that their sensor data is insufficient to perform the tasks, and the recorded data won't be usable for training a system with whatever sensor is needed; in which case, no advantage for logging customer movements for N years.

> all it takes for Tesla’s NN approach to pay off is one leap in explainability.

It's not vision and disengagement-led training that's conferring the scale advantages you're talking about, because other AV companies actually do train on and attempt to drive in various kinds of weather, and safety drivers do the same thing normal drivers do except that they're more trained. The difference is the approach Tesla has to safety (vs the others). Which is to say: Tesla doesn't worry about training their supervising drivers, at the risk of those folks' lives.

Tesla also doesn't seem to worry about _proving_ the safety of their systems: explainability isn't simply going to be a matter of a new algorithmic breakthrough. How do you prove that your software is safe if you don't sample heavily from different operating domains you want your car to work in, and then either simulate on those miles or actually test your car? Before letting a driver behind the wheel of an autopilot, companies like Cruise/Waymo/Aurora all do manual driving first in a new domain, and then move to safety drivers (even 2 to start, for extra safety). This is not because their software is incapable of doing the same things as Tesla, but because it's very worthwhile to sample disengage rates and assess safety criticality to various issues before allowing the car to act autonomously. And even then there are safety nets in place to further mitigate risk. (You're probably right that L5 is a long ways off, but this is how we'll get to L4.)

Of course it's more "scalable" to avoid all this and put the responsibility of risk on users who, as your system improves, become less and less attentive. And so this is what Tesla does. I suspect solving the long tail of issues is going to be more of a challenge for Tesla because at least in my experience it's easier to {convert a naive rules/simple model --> complex model} than to {debug the complex model --> handle failure modes with logic}.

In self-driving, you basically have 3 parts - perception, prediction and planning. Perception is the easy part (if you have lidars). Tesla and AFAIK others rely heavily on NNs for perception. The real challenge is prediction and planning. To what extent are NNs involved here in Tesla's system is unknown, probably not that much. Overall, I don't think that Tesla has a fundamentally different approach than others.

The moment in the video just after 7m0s seems to contradict this.

So they are almost there with the part others barely even have to do, while the part everyone else is working on will eb ridiculously good once they get round to it?

I'm skeptical.

Or is the software just hardcoded to assume that if it turns the steering wheel at a specific angle that the car will move a specific amount in that direction?

The Model 3 and Y have only the center screen in a horizontal layout.

https://tesla-cdn.thron.com/delivery/public/image/tesla/2558...