[gambiting]

Yes, of course Tesla hasn't solved it yet, because no one has, and in fact it probably isn't possible to solve without some magic tech that doesn't exist yet - it might be possible to fix with LIDAR, but obviously Tesla doesn't have it.

All of these systems have to ignore stationary objects ahead of them, or the car would be emergency braking in too many daily situations to be usable.

Simple case, at 150km/h you are travelling at 41m/s. The stopping distance at that speed is roughly given as 130m[0]. So the car would need to see and recognize objects at least ~150m ahead to stop autonomously from that speed. That's just physically not possible, the cameras on the Tesla don't have enough resolution to do such a recognition. Instead, a radar-based distance measure is used - but again, even if the car detects that you are rapidly approaching "something" 150m ahead of you, that information is nearly useless. At that distance, you cannot differentiate between a stopped car 150m ahead of you, an overhead sign, or a large rock next to your lane which poses absolutely no danger whatsoever - it all gives the same signature. LIDAR doesn't have that range either. And then of course even if the car could reasonably detect that there is something ahead of you that you are absolutely 100% definitely heading towards, it has no idea if the road doesn't curve in such a way that you would avoid it. Famous case of adaptive cruise systems freaking out at bends, because according to the radar/image recognition you are CLEARLY heading for that telephone pole standing next to the road - but of course the road curves so you aren't actually going to hit it. Problems like that.

[0] https://www.random-science-tools.com/physics/stopping-distan...

[diggernet]

Sorry, coming in late, but gotta object here...

My Subaru has no problem detecting stationary objects. It uses stereo cameras, for which object detection and locating the object is a thoroughly solved problem, whether the object is moving or stationary. No magic needed. It just works. For example, there is a sharp turn near my home with prominent turn signs along the curve, which puts them directly in front of you as you approach the turn. Those stationary signs quite reliably set off my collision warning if I'm approaching them too fast.

Tesla has the problem because it uses radar for object detection. The radar can tell them the presence of an object, but cannot tell them the object's location. An overhead sign or overpass or whatever looks the same as an object in the road. So to avoid countless false positives, they need to ignore signals from stationary objects because they chose to use radar. It's a self-inflicted problem that other vehicles do not have.

[jessaustin]

Aren't there fully self-driving vehicles undergoing testing on normal city streets with stoplights? We could say that's a different mode than high-speed highway driving, but there are a number of locations where those modes blend into each other. Are the experimenters extra careful to avoid those locations? How long before the tech exists to address this issue?

[gambiting]

Sure there are, and yes, they are different sets of problems though. Have a look at the British Tesla Driver Youtube channel, some of his videos are eye opening. Basically the car is in full autopilot mode, approaches an intersection, correctly slows down, waits for its turn, starts moving.....and in the middle of the turn goes BEEP BEEP BEEP and disengages entirely because the road markings aren't there and it wasn't entirely sure where to go. And now of course you're in a moving vehicle that's heading for a collision with someone else and requires IMMEDIATE attention to continue. One could(and I'm sure will) argue that the system "shouldn't be used this way". But that's a moot point, if the system is there and lets you do this, then people will use it this way.

"How long before the tech exists to address this issue?" I'm not sure if that's a problem with tech as such. We have fantastic cameras, yet famously Google's best image recognition algorithm just couple years ago would reply, with 100% confidence, that a sofa in a zebra print is in fact a Zebra, after all the stripes are there, it has 4 legs.....it must be a zebra.

So in my(personal) opinion, self driving will face the same challanges image recognition has faced - we will rapidly get 90% of it right, then the last 10% will be a massive pain to get right for decades if ever.

[Isinlor]

It's an active research field. E.g. from October 2020: Calibrating Deep Neural Networks using Focal Loss: https://arxiv.org/abs/2002.09437

> Miscalibration - a mismatch between a model's confidence and its correctness - of Deep Neural Networks (DNNs) makes their predictions hard to rely on. Ideally, we want networks to be accurate, calibrated and confident. We show that, as opposed to the standard cross-entropy loss, focal loss [Lin et. al., 2017] allows us to learn models that are already very well calibrated. When combined with temperature scaling, whilst preserving accuracy, it yields state-of-the-art calibrated models. We provide a thorough analysis of the factors causing miscalibration, and use the insights we glean from this to justify the empirically excellent performance of focal loss. To facilitate the use of focal loss in practice, we also provide a principled approach to automatically select the hyperparameter involved in the loss function. We perform extensive experiments on a variety of computer vision and NLP datasets, and with a wide variety of network architectures, and show that our approach achieves state-of-the-art calibration without compromising on accuracy in almost all cases.

Calibration will be practically solved in couple of years. Then a bit longer for addressing adversarial robustness.