my friend bought a ride from a waymo for shits and giggles. i was actually intrigued about how the car made decisions on its own and sorta nerd sniped me to learn more about machine learning (not llms).
there’s probably some evil thing the self driving car companies are doing that im not aware of, but I was just interested in the ML aspect.
I just want busses and trains.
I’d rather have a heads up display
They will be the future. Not today, not tomorrow but some day. When they have less accidents, especially deadly ones, you have a hard time to argue why human drivers shouldn’t be banned. Of course there are still a lot of things to handle and to figuring out. One of the biggest causes of accidents are humans. If nobody can drive, you eliminate a big part but there is still humans on the road, being it pedestrians or bikes. If you separate them from the road, this is a huge step. We would need to do a lot of work, making the streets good for AI cars. We also need a lot of discipline, to keep the streets safe for AI. Being it to maintain public available maps and directly update them or discipline with temporary markings. I’ve seen quite a few examples, where the signage was wrong and if I would have followed it, as the law said, I would have driven wrong and especially bad, other cars around me, would have had other signs. This wouldn’t work with AI cars, as they can and should follow strict rules and allowing them too much rule breaking, would be bad.
I think what we would see first is a hybrid system, where certain routes can only be driven by AI, like highways, where its easier to maintain them for AI. Cities would highly benefit from it, but I think its quite a challenge to optimize it.
Next thing where AI cars would be better would be easier flow of traffic, no traffic jams and less need for an own car and parking spaces. If all cars communicate with each other, they can accelerate all at the same time, which massively boosts the throughput of a street. Also just for cars, there is no need of traffic lights. They communicate between each other and coordinate how they drive across a crossroad. As for the other thing, cars could just drive around like taxis. Without a human driver, they can drive around all day, are way cheaper and don’t need a parking spot everywhere. You wouldn’t need parking spots in cities at all. You would only need spaces for cars to stop, to pick up and let out guests and somewhere a charging spot, if we don’t come up with a system that charges them, while driving. But this charging spot could be anywhere and they need way less space, as they don’t need to open the doors.
You may say, this is in the far future, but let’s remember human inventions. 1902 the Wright Glider took it’s first flight. Just 67 later in 1969, we send the first man to the moon. Today we wear watches, on our wrists, that are 50,000 times more powerful, in some specs even more than that, than the Apollo 11 computer. 100 years ago we didn’t had televisions. In 1935 we had the first TV station. It would be bald to assume, that it would take ages, until self driving cars would be ready to fully drive alone 24/7. I would assume that I would still see the beginning during my lifetime, where on certain routes, no human driver is allowed. Just look how far AI video gen has come. Take the “Will Smith eating spaghetti” gen Video. It’s wild how far it had come in just a few years.
Of course today they aren’t ready and I would be pretty scared, behind the wheel. I didn’t had the “joy” to test this, but I absolutely will, when I get the chance, just to see how the tech is. the whole testing in a real environment is needed, to see the weakpoints, but I think they went too fast, especially when I see videos of cars, deadlocking themselves. This way we do get the data quicker but sometimes at a high price.
I hate driving and like the idea of driverless cars, but I’d far more like to have better mass transit and more areas that are enjoyable to walk around in. More space for humans and nature, less for roads and parking.
Why do they not have to pass the same test as every other driver on the road? For every iteration of the software they use?
Are you sure they don’t already have more rigorous controls? I took my driving exam in a small town middle of the nowhere and can now legally drive anywhere, even major cities with traffic.
I would never trust a self driving car which passed ONLY that exam.
They fail basic driving skills that would fail a student driver, and are immune to the coercive controls that other drivers face, such as fines, loss of license, and jail time.
And would you like to make a comparison to people? I see people do shit that would “fail basic driving skills” every goddamned day, and sometimes its me doing them.
Given how bad humans are at maintaining attention, the standard that self-driving be perfect is a little silly. The standards debate should START at “no worse than the average driver.”
I agree with the comparison to people and how poorly people drive.
There is probably a lot of confirmation bias in my low opinion of self driving cars, because we don’t have them here and the only stories that make the news are failures.
Quotes from CEO’s saying that keeping them out of bus lanes is too hard don’t help. Nor do self aggrandising lies about their abilities.
They do have a lower accident/harm rate than human drivers, but my point of view comes from a c/fuckcars perspective. I live in a car centric society, and have three and need a fourth (5 people, 3 teenagers going in different directions) which is fucking ridiculous. I would rather see public transport be reliable and accessible.
My city has made steps forward recently; all public transport is 50cents, all the time. The problem is we have to drive 20-35 minutes to park, to use the public transport.
I think Uber under pays me and other drivers to this extent:
But has $10 billion to invest in driverless vehicles to replace us rather than paying decently.
So…meh. Not a huge fan.
Did you know that the less picky you are on choosing your routes (or in other words, you are desperate for money) the less they pay you? Great business
damn, that’s probably the evil part i was yet to learn about
I fucking despise them. They don’t solve any problems. They’re dangerous as objects on the road. They can get stuck in loops. They are ridiculously expensive to maintain. They are an integrated surveillance system. They don’t even know how to truly drive; Waymo cheats by giving them maps, and the second that map no longer applies they fail miserably. They’re dangerous to construction workers. They enable morons to drive and get away with it for long periods of time before disaster strikes, and when it strikes, someone else dies not the moron in the “driver’s” seat.
Self driving cars are a replacement for classic public transport that is worse in every conceivable way.
I think they are wasted time, money and energy. They are a solution to problems that have been solved long ago in much better ways: better urban planning and public transportation.
They’re a distraction. We don’t need self driving cars, we need trains and reliable busses.
I don’t trust them.
I feel like with most cases of post 2008 technology, what they’re doing is interesting on the surface level. The problem is that it’s always a wildly inefficient and predatory solution for whatever problem.
Self driving cars sound like a good idea, smoother traffic, safer roads, etc. The problem is the technology isn’t really there yet and the way it’s currently implemented is dangerous and inefficient and really has no place on actual roads. So what we end up with are waymos that are error prone and susceptible to harassment from people and teslas causing accidents. That harassment waymos get is somewhat justified too because of what they represent. Instead of billions of dollars being spent on public transit which would solve the same problems more efficiently, billions of dollars are spent on a personal, luxury, novelty transit experience.
As far as the ML goes, yeah it’s interesting I guess. It needs vastly more data than it has because driving is incredibly complicated. I don’t really have any basis for it, but I kinda don’t think it’ll ever be effective enough until we reach some kind of technology breakthrough with quantum computing or something. There’s too many data bottlenecks for the amount of data processing required. Until we get that, waymo is just gonna have to keep relying on underpaid Filipino support staff to fill in the gaps.
-
In order for these vehicles to operate, they must all be surveillance platforms. They are covered in cameras, microphones, radar, sonar, lidar, and other types of sensors. The more of these vehicles are on the roads around you, the more real-time surveillance data is being collected. And you better believe those cameras are doing license plate recognition of every other vehicle they pass, in case there’s any incident that would involve an insurance claim. Basically they are mobile Flock on steroids.
-
Training machine learning models requires data. The edge cases have the least data for training, because they involve unusual weather or road conditions, specific lighting situations, weird/poorly designed intersections, or other uncommon circumstances. Edge cases will never be trained out because there will never be enough data for each specific case to have a meaningful effect on the self-driving model. Edge cases present the highest likelihood of injury to humans, whether as passengers, other drivers, or bystanders. People will continue to die, and somewhere in a meeting room someone will be presenting calculations to a board about whether it’s worth investing in better safety controls for their vehicles, or just bribing politicians to shield them from regulations and lawsuits.
People will continue to die
People also die in droves daily with human drivers. The important point is whether self driving cars are safer than the average human, not whether they eliminate all accidents entirely.
-
Waymo uses predefined driving maps so it’s a lot more reliable, or supposed to be… this is how it probably should be done.
Upper echelon on YouTube has a really great video on accident and reliability numbers from both waymo and Tesla. Tesla is shockingly bad in safety. I highly recommend the watch, it shows a serious level of fraud.
Self driving is great and all, but it’s kind of fixing a problem that should be fixed by better city planning and public transport
Ooh, I get to geek out for a bit!
I helped a friend get into waymo as an employee when it was still staff-only/non-public, and later got to attend a private event where I could explore the regional hq building freely, view the cars, the hardware that makes them go, the nav system of all the cars in the region in real-time, see video feeds and diagnostics in real-time, ride in a ‘beta’ style vehicle with wires and sensors very much visible, ask questions directly to technicians regarding details about the software, cameras, lidar, emergency situations… The whole nine yards. I have also been in a public vehicle that was able to override the geofencing and ‘guardrails’ for a couple trips as well, and see how that works, how the car communicates with hq when that happens (‘why are you doing this, is everything okay, etc’). Except for someone currently or formerly employed by waymo, I am possibly one of the most informed people about how it all works. Especially the Lexus and Chrysler vehicles, since I got special access to those.
That out of the way, you are correct. They run the cars with a safety driver while training/feeding data back to the regional hq while in a new area, so you might see cars with a person at the wheel (and possibly a passenger, sometimes with a laptop) while the car drives around, not stopping for passengers. They note issues and areas for potential problems, log many miles along the same roads at different times of day for an entire area, and will do this with multiple vehicles for 6-12 months, before having enough data, runs, low incident occurance, and high confidence before allowing testers to ride; my last information was that they still retain the safety drivers for several additional months, before letting the cars go ‘solo’ and open the area to the public.
The friend has moved to greener pastures but this data is between '15 and '23, I believe. I am signed up for the tester program but the service is not yet in my home area; I’m a few miles outside their current zone for my region. I’ll be in the service area this winter though, and am planning on taking a couple rides and refreshing myself on the current situation for the general public.
I’ve never seen a Waymo in person but I’ve seen plenty of edge cases while driving that I wonder how any self driving car would handle it.
Just one recent example: I was at a red traffic light crossing another road that had virtually no visibility to the right due to thick vegetation. Just as the light turned green I heard the sirens of approaching fire trucks from the direction where I couldn’t see. I waited, and didn’t actually see them until they were about 20 feet from the intersection. My gut feeling is that pretty much any self driving car would have seen the green light and then driven right into the path of those approaching emergency vehicles. And what would have happened next? Would the self driving car have slammed on the brakes once it saw the fire trucks? Would it have just continued, forcing the fire trucks to stop?
I’d rather have trains and walkable cities
