I think the argument is that if a human with two eyeballs can do it, a computer with a bunch of cameras should be able to as well. I can drive in the snow.
More often than not, snow driving is CHALLENGING. And it's not for a lack of experience, as I'm doing multiple roadtrips through the BC Coquihalla Highway. More often than now, it's more of a matter of best attempt at driving in the path made by other car disregarding lines on the road than driving by the book. The utter HELL being driving at night with a blizzard, this is a level-up experience for sure.
1. Many people have no choice but to drive in bad snow conditions for any number of reasons including that they're emergency workers.
2.) Winter conditions catch you. Even though I don't commute and basically do everything to avoid bad winter weather conditions, I've still been caught in pretty bad conditions a couple times over the past few years.
Fuck them...
Is that how much a pussy society we have become ? Avoid dangerous conditions because "it might be dangerous" ? If a freaking Tesla can't handle what my 06 SUV can handle without issue, and what people were handling ~forever, ... just as much as an Apple laptop can't handle what a then-IBM was handling, then the damn "car"/"laptop" is nothing but a piece of shit.
Urbanus homo sapiens sapiens is really a pussy...
A human can drive in the snow, and has a sensing package which effectively consists of two cameras.
The hard part (when you only have two cameras) is that the processor needs to be exceedingly sophisticated, e.g. a human brain.
People like Waymo have decided the best way to decrease the brain requirement is to add more sophisticated sensing that humans are fundamentally incapable of.
Tesla's strategy is more to just build a better digital driving brain with machine learning + lots of miles, because arguably all sensing mechanisms have their shortcomings. That being said, they still have a lot more than two cameras in their FSD package – just not LIDAR. They still have radar and other things as well.
Only time will tell which approach is better. Probably in the end the best approach will be to combine the two: a brain that is sophisticated enough it can drive with just cameras, but you throw LIDAR and such in their too to go the extra mile.
Two cameras, and a huge amount of mostly not understood wetware interpreting their signals. Our current approaches to ML may be a local maxima. The true path to AGI be somewhere as yet undiscovered. Lidar certainly seems like a good tool to use given our current ML capabilities.
No LIDAR means more training data, as has been made clear by the amount of miles driven by Waymo vehicles vs Teslas in Autopilot.
A generally intelligent driving system is going to benefit from better sensory inputs just like a human would, it's not like that makes the system dumber. The whole dichotomy between sensory inputs and processing seems strange to begin with, there's not necessary a clear line, or a reason why one should be prioritized over the other.
You have to realize that most SOTA self-driving systems are using a lot of NNs. NNs in turn want lots of training data. That training is not fully generic – in other words, if you train a NN using only cameras, giving LIDAR to that trained network (i.e. for inference) will not really do anything.
Having large LIDAR arrays on the top of vehicles is expensive and slow to deploy, which directly translates to being able to generate less training data.
So if your goal is to have the best possible self-driving vehicles 100 years from now, absolutely you should be using as much sophisticated sensing equipment you can get your hands on, and slowly train your nets for the next century. If your goal is to get out something that is better than a human driver as fast as possible, then it is not clear that requiring all vehicles to have LIDAR on them is the best way to go.
The argument for having a lidar vs not having a lidar is different. They chose to go without lidar because it should be cheaper and faster to get started, and the fact that people are capable of doing driving without lidar makes it seem possible that a machine could do it without lidar.
Keeping the context of a self-driving car, humans are perfectly capable of driving cars with just two eyes to determine surroundings. For 99.9% of your driving, you don't have to think about it. Stay between the lines. Stop for the red light. Make sure there's nobody next to you before changing lanes. There's no creativity here, just following a very concrete procedure.
Humans fail because humans do stupid human things. They text while driving and drift over the lines or don't see the red light. They don't always check before changing lanes. They drive drunk. Computers won't do any of these things.
And they'll do it better. Humans have only two eyes and can only look in one direction. A computer can check the blind spot while still looking ahead, so there's no chance it will rear-end someone because the person in front of them slammed their brakes while it was making sure the next lane was clear before making a lane change. A computer will notice a car in the blind spot veering into its lane without taking attention off the road in front.
Again though, it's all down to the software. We're not there yet. But that doesn't mean we can't be in the future.
On the contrary, having computer vision capable of handling snow is a necessary precondition for self-driving, regardless of what other sensors are present.
I think this is a huge simplification of the actual problem. Not only do you have to denoise the data, you have to use the data to generate models of 3D objects in space. Maybe you could argue that exact 3D models are not necessary, but IMO any neural network that can drive at Level 4/5 needs to approximate 3D space in some layer.
This is incredibly difficult; we have an entire region of the brain dedicated to for this exact purpose (the visual cortex). This region has the benefit of millions of years of evolution. We also have the advantage that we can predict 3D models based on everyday experience; we don't know how to encode that knowledge into a neural network efficiently. That's not even mentioning the latency requirements that driving in the real world would impose on such a network.
The difficulty imposed by snowfall itself is just de-noising.