Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Waymo pauses Atlanta service as its robotaxis keep driving into floods

techcrunch.com

Post reply on HN
Print view View original

91–100 of 498 posts

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#91

Post by michaelt »

moomoo11 wrote: post #11
colordrops wrote: post #10

Self driving will never handle all corner cases until they essentially have a frontal cortex. They probably need something like an LLM to help with very high level abstract situations, e.g. avoiding a hurricane like someone else mentioned in this thread.

how would a llm help maybe a little biological brain engineered to think it is a car with api access to the car hardware via the llm? imagine you get into the car and in the center console you just see a floating brain in vat like fallout

The driving ML model will take care of the next 10 seconds of driving, in a fast loop deciding what steering and throttle commands to give.

The LLM will apply the high level reasoning needed to deal with longer time horizons and complex decisions, like deciding that the best way to reach the car wash 100 yards away is by walking.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#92

Post by wongarsu »

thewebguyd wrote: post #43
xnx wrote: post #6

I wonder how much of this is trouble perceiving water depth vs integrating that understanding into the larger driver model without creating regressions elsewhere.

I don't think there's a good solution right now. You can't just go based on surrounding traffic because humans are also stupid and flood their cars all the time. You could maybe use short-wave infrared cameras combined with ground penetrating radar, but it'll get real expensive so probably not commercially viable. I think the only "good" solution is to have the car be overly paranoid, and if it detects water on the r…

Their fleet is constantly scanning the area with lidar, which is assembled into maps. If those maps are in 3d rather than a 2d road grid you can calculate puddles very accurately with no extra sensors:

- Find the edge of the water using vision or lidar

- look up the ground height at that position in your map data. That is the water level

- run a flood fill of the local 3d map starting from that point, with that water level. That gives you an exact shape of the puddle

- for any point on your planned path, you can now check if the point is in the puddle (per the flood fill above) and how deep the water is (difference between puddle's water level and ground height)

- use that either as a go/no-go for a planned path, or even feed this into your pathfinding to find a path with acceptable water level

The main limitation is that it assumes that the ground hasn't changed. It won't help in a landslide, or on muddy ground where other cars have disturbed the ground. But for the classic case of the flooded underpass or flooded dip in the road it should be very accurate

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#93

Post by retrocryptid »

paxys wrote: post #13

Driving through an obviously flooded street thinking "I'll easily make it" and getting stuck in the middle? Yeah, these cars have achieved human level intelligence.

That being said... it's actually somewhat uncommon for humans to drive into flooded streets. To the degree that people think it's notable enough to take videos and post them to social media. I don't have the data, but would be interested to see how many times per passenger mile travelled human-directed and remotely-operated vehicles like Weymos drove into flooded streets.

I can appreciate the cameras and lidar on the Weymos don't give their remote operators a lot of good data about the depth of water on the road-way. As you point out, humans in cars often don't get this right. I think the humans that don't drive into deep water are the ones who a) give any amount of water on the roadway a big NOPE and b) people familiar with the local environment and use multiple visual clues to judge the true depth of the flooding.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#94

Post by etskinner »

b40d-48b2-979e wrote: post #39
loudmax wrote: post #24

Earlier quoted context omitted.

Ideally, robot drivers will some day be better drivers than humans in all road conditions. They'll be able to coordinate fast lane merges and busy intersections by subtly adjusting speed without vehicles having to stop. Imagine a busy intersection where all the cars fly past one another at 40 miles an hour without stopping but none of them crash. Humans can't do this, but machines could , if, and when the technology…

Evidence suggests... no, that day is never coming.

Once all cars are autonomous, that day is certainly coming. Even before then, it's very likely we'll see platooning in the future, even if there are still some human drivers.

Also, this already exists in some places. Look at a video of how to cross the street as a pedestrian in Vietnam: You literally just start walking across and people weave around you. Or look at driving in India and similar places.

All I'm saying is never say never

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#95

Post by stetrain »

losvedir wrote: post #88

I think another way of framing it is "Waymo pauses Atlanta service due to weather conditions", which doesn't sound at all unreasonable to me. It's no different from "Chicago O'Hare pauses flight departures due to a winter storm" or whatever. I think that self driving cars won't ever be able to handle every condition out there, and so there's probably a time when the system will be paused / shutdown when conditions ar…

Well except that there were incidents of cars getting stuck in floods with passengers before they paused the service.

A closer analogy would be ""Chicago O'Hare pauses flight departures due to a winter storm after 3 planes slide off the runway due to ice"

Absolutely I think there will be a disconnect between when people think they should be able to drive somewhere (ie to work in a no-visibility blizzard) and when ideal self-driving cars would allow themselves to operate. Maybe society will adjust to be more flexible to natural conditions, or maybe people will get frustrated and drive themselves into the poor conditions as always.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#96

Post by CSMastermind »

mschulkind wrote: post #84
fastball wrote: post #72

Earlier quoted context omitted.

This is why I personally feel like Tesla's approach is more likely to "win". The fundamental blocker to self-driving cars is not sensing / sensor fusion, it is intelligence. And the Tesla approach seems much more likely to achieve functional intelligence than Waymo's.

While I agree with basically all of this, and find the FSD on my Tesla to be quite useful, a question pops into my mind. Why can't Waymo ALSO develop the same smarts and just also solve the sensor fusion issue such that they can use the right set of sensors in the right environmental conditions, and then leapfrog Tesla's capabilities?

I got downvoted for saying this last time the topic came up but constraints focus a project. It’s best to start work with as few variables as possible, and only add new ones when absolutely necessary.

I'm working on a similar problem in computer vision and we're quickly approaching the point where our pure vision work is better than our Lidar supported track because we've had to deal with the constraints instead of having a crutch to lean on.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#97

Post by eptcyka »

loudmax wrote: post #54
colordrops wrote: post #10

Self driving will never handle all corner cases until they essentially have a frontal cortex. They probably need something like an LLM to help with very high level abstract situations, e.g. avoiding a hurricane like someone else mentioned in this thread.

Humans don't handle all corner cases. People can be slow to react to completely novel or surprising situations. There will be corner cases where humans generally do better than a machine, but the simple rule to slow down and come to a halt if things look too weird or confusing will almost always be the right answer. Ideally, driverless cars will one day be better drivers than humans and this will save tens of thousan…

I think self-driving cars will only become better once they can do all the learning in real time and on-board. Otherwise, they will only be as good as the data they trained on - which is ultimately real meat driver data and a derivations of said data.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#98

Post by ai-x »

mschulkind wrote: post #84
fastball wrote: post #72

Earlier quoted context omitted.

This is why I personally feel like Tesla's approach is more likely to "win". The fundamental blocker to self-driving cars is not sensing / sensor fusion, it is intelligence. And the Tesla approach seems much more likely to achieve functional intelligence than Waymo's.

While I agree with basically all of this, and find the FSD on my Tesla to be quite useful, a question pops into my mind. Why can't Waymo ALSO develop the same smarts and just also solve the sensor fusion issue such that they can use the right set of sensors in the right environmental conditions, and then leapfrog Tesla's capabilities?

I thought about this and I think it boils to how the model is trained.

Tesla trains it models from actual drivers purely based on (input) Vision and (output) actuators - Brake, Steering, Accelerators.

Human output is based on what they and the camera sees. So, it's a 1:1 match.

If Waymo were to do that, it'll muddle the training set. The Lidar input may override camera input.

I always struggled when Musk mentioned Lidar will make it ambiguous. It didn't make any sense to me why having a secondary failback sensor messes things. But, if you put it in the training data context, it absolutely makes sense.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#99

Post by briandw »

mschulkind wrote: post #84
fastball wrote: post #72

Earlier quoted context omitted.

This is why I personally feel like Tesla's approach is more likely to "win". The fundamental blocker to self-driving cars is not sensing / sensor fusion, it is intelligence. And the Tesla approach seems much more likely to achieve functional intelligence than Waymo's.

While I agree with basically all of this, and find the FSD on my Tesla to be quite useful, a question pops into my mind. Why can't Waymo ALSO develop the same smarts and just also solve the sensor fusion issue such that they can use the right set of sensors in the right environmental conditions, and then leapfrog Tesla's capabilities?

Because they don't have a fleet of millions of people labeling the data for them and paying for the privilege of doing so. Waymo has about 3700 vehicles. Tesla has millions. Waymo only operates in known environments and collects a very limited range of data. Tesla collects data everywhere that people drive their cars.

Re: Waymo pauses Atlanta service as its robotaxis keep driving into floods

#100

Post by tptacek »

etempleton wrote: post #90

This is really my bear case against AI. I am not against it. I actually think it is really neat! But we have been working on driverless cars for how long and spent how much? And still things like a flooded roadway completely throw them. Tesla failed to deliver driverless cars but now is pivoting to the much more complex fully autonomous robots. And we can’t get AI to stop hallucinating facts, but any day we are going…

I was (I think the search bar will prove this out) a pretty committed skeptic of driverless cars, but I've come around on them in some use cases. I'm not optimistic about them on highways. But they solve some important problems in regional/local transit.

We're contemplating standing up an EV shuttle service in Oak Park. It will fail. As I understand it, we've piloted non-EV versions of a shuttle service; they failed. The problem is that in small local areas, the staffing for a useful transit service is too expensive; that's because "useful" imposes constraints about responsiveness, coverage, and most of all hours of service, which mean the service won't pencil out with the ridership it'll get.

An autonomous vehicle transit service in our muni would probably work fine; it's a strict grid system with very low speed limits (AVs will, in our area, be strictly better drivers than the median human drivers --- this isn't a statement about human fallibility so much as an observation about scofflawry in our area). And if the product existed, we could afford it, because we wouldn't be paying fully loaded headcount costs for 2+ shifts of drivers at epsilon levels of utilization.

For whatever it's worth, I don't really have "autonomous vehicles" and "LLMs" in the same bucket in my head. I'm bullish on both, but for very different reasons. It usually doesn't occur to me to think of Waymos as "AI", though, obviously, they are.

Post reply on HN