And this is the 'hard bit' about self driving cars. If they are too passive and cautious human drivers will exploit them, if they assume the other cars will follow the rules, they will hit them.No amount of LIDAR magic and deep learning can fix that problem yet.
Is this going to be inherent and unavoidable for the machine learning based self driving cars? Perhaps self driving is never going to become a reality, just because of the problems like this (unless/until the road infrastructure supports them explicitly).
Right now, per mile, self-driving cars are much less safe than humans. As has been noted before though, we don't have that many miles of data on self-driving cars, and the tech is still quite young and there's (probably) a lot of room for improvement.
The issue is that self driving needs to be better than human drivers by a considerable margin. The public will not tolerate an AI driver that makes the same mistakes as humans, they expect better. To do this it will need access to sensors that humans don't have
Human drivers already space out, check their Facebook, text their friends, and otherwise lose track of what's going on without having cars that claim to do most of the driving autonomously. The more autonomy you add, the more drivers will take advantage of the opportunity to do something behind the wheel besides drive.
Autonomous cars are one of those things that really need to be done right or not at all. The notion that human drivers will be ready and able to take control at a moment's notice is nothing short of laughable.
IMO, about the only way partial self-driving cars can work would be to delegate it to a remote building full of 'drone drivers' somewhere. Heck, I'd pay for that today, just out of laziness.
This kind of thing is why I think we are much farther away from true L5 self-driving than many others think. To handle current road conditions you need true AGI with the ability not just to see as well as humans can but to interpret what you see, as well as theory of mind and contextual awareness of the drivers around you.
What I think will ultimately happen instead is we’ll adapt the roads to allow self-driving by dumber computers. For example, a self-driving lane on highways where there are sensors on the road to help the cars, and where all the cars in the lane must be self-driving and in communication with each other to do things like maintain stopping distance.
"Self-driving cars today work, and are cost-effective, and it's likely that they will be safer and create a much better traffic ecosystem than humans ever will."
Self-driving cars ONLY work on the immaculate freeways of California that they are almost always (unfairly) tested on. Any precipitation renders the LADAR sensors all but useless, and many everyday driving scenarios like "left turn onto traffic" and "no lane markers" are still are far from being solved yet.
Any real improvements in the technology will be the result of fundamentally different techniques than what the state-of-the-art is currently using (since this is Hacker News: state-of-the-art really is just an "ad hoc" pipeline that looks for things like "lane markers" using things like Canny edges along with a PID controller for the steering wheel actuator, with some other "cheats" like an over-reliance on human-compiled map data provided "a priori").
An end-to-end deep learning approach seems promising, but the current results aren't even usable at this point.
Five years from now, every car (even my Jetta, not just luxury cars) will have the equivalent of what Tesla's "auto pilot" looks like now, but a human in the front seat will still very much be a necessity.
Here is my proposal for self driving cars: forget trying to make them drive on local roads at first. Add transponders to all interstate tarmac that can be used to detect location and lane position. Then use a much cheaper LIDAR just to map where the other cars are. Boom: you have made long haul driving autonomous. You can then slowly expand this to smaller roads. But why try to recreate human drivers when machines can use much better sensors than us, but can process visual info much slower, and gather it with worse fidelity?
1) We don't just rely on our vision when driving but also on sound (nearby cars) and touch (feel of the road). I actually haven't seen any self driving car projects talk about this aspect which is interesting.
2) Self driving cars can't be just as good as humans they need to be effectively perfect. Quite a lot of people think they are amazing drivers i.e. infallible so mistakes from a self driving car aren't going to go down well.
I have been saying this for years. Even if self driving cars handle 99% of situations correctly, they don’t become feasible until they can do that last 1%, which is 100x more difficult then previous 99. The reality is, if the car drives itself, people will be doing other things in the car and will be unable to respond to emergencies.
Kind of interesting to consider how this adjusts priors on these outcomes:
- Self driving is not possible
- Self driving is not possible anytime soon
- Self driving is possible, but requires LIDAR
- Self driving is possible, and can be done with normal cameras
I've engaged with a lot of people who presume self-driving reduces onto AGI, therefore it will not be achieved anytime soon if ever. I wonder which of their assumptions is wrong, if this ends up being successful.
The main problem is totally technical! Robot cars can barely deal with following white lines on the road. How about rain, construction, poorly marked roads, fallen trees, badly parked delivery trucks, emergency vehicles, pedestrians, cyclists, etc? Can cars read roads signs and understand them? We'll have full-on AI before self-driving cars are for real.
Interesting as something to explore, but I'm curious if a network of self-driving cars can be optimized to recognize and avoid these situations altogether.
Part of self driving is dealing with those human drivers who are terrible drivers though.
This is what makes the whole thing so hard. If we could have dedicated roads for self driving cars it would be easy peasy. No need to interpret road markings, signs, other vehicles and their intentions. Because it would all be communicated digitally over radio. Steering a car is total peanuts for a computer.
It's the unexpected stuff that makes it hard. Like that Uber car that killed that lady that crossed the road with her bike in the dark at a totally unexpected spot. A human is better at those things because they're caused by other humans who aren't logical.
But such matters are really hard to test and evaluate for precisely because they are unexpected.
Safe self driving cars without some kind of outside coordination with either the road or other vehicles is an impossibility. It's impossible to do this safely on a mass scale without the road infrastructure made for it.
If the technology were as undeterministic as this post makes it seem to be, there should not be any self driving car allowed even under the best conditions.
I fully agree that an AI can be fooled, but that is high level logic (path planing), the system should be designed to have a fallback that does emergency braking if all else fails.
There simply is a point where the high-level AI does not matter any more. And that is if I (the car) am moving at 45 mph towards an obstacle that is in the middle of the road less than 2 meters from my projected path.
This does not mean that a full brake is required but the speed definitely needs to be reduced to account for the uncertainty, and once it is determined that it is physically impossible to miss the obstacle the system must do a full stop to reduce the impact velocity as much as possible.
It's fine, if the LIDAR data is plugged into a machine learning algorithm, and you will probably get less than the 10-20 Hz the scanner can produce, but at the same time it is probably also used by a much simpler obstacle tracking algorithm that can run at near real-time speeds.
The self driving car doesn't need to be perfect, it needs to be better than humans. That is a much lower bar. Self driving cars have already demonstrated their ability to handle traffic better than humans, but there are other situations where they are much worse.
reply