Waymo robotaxis are now giving rides on freeways in LA, SF and Phoenix

This is a huge sign of confidence that they think they can do this safely and at scale... Freeways might appear "easy" on the surface, but there are all sorts of long tail edge-cases that make them insanely tricky to do confidently without a driver. This will unlock a lot for them with all of the smaller US cities (where highways are essential) they've announced plans for over the next year or so.

Looks like they've opened up SJC Airport, too! SFO imminent?

Freeways are easier than surface streets. The reason they held off allowing highways is because Waymo wants to minimize the probability of death for PR purposes. They figure they can get away with a lot of wrecks as long as they don't kill people.

There's also the risk of a phantom breaking event causing a big pileup. The PR of a Waymo causing a large cascading accident would be horrible.

Waymo (prev. Chauffeur) were cruising freeways long before they were doing city streets. Problem was that you can't do revenue autonomous service with freeway-only driving.

The real reason I see for not running freeways until now is that the physical operational domain of for street-level autonomous operations was not large enough to warrant validating highway driving to their current standard.

"Easier" is probably the right one-word generalization, but worth noting that there are quite different challenges. Stopping distance is substantially greater, so "dead halt" isn't as much of a panacea as it is in dense city environments. And you need to have good perception of things further away, especially in front of you, which affects the sensors you use.

> Freeways might appear "easy" on the surface, but there are all sorts of long tail edge-cases that make them insanely tricky to do confidently without a driver

Maybe my memory is failing me, but I seem to remember people saying the exact opposite here on HN when Tesla first announced/showed off their "self-driving but not really self-driving" features, saying it'll be very easy to get working on the highways, but then everything else is the tricky stuff.

Also on surface roads you can basically stop in the middle of the street and be annoying but not particularly dangerous. You can’t just stop safely dead in the middle of a freeway.

the difficult part of the highways is the interchanges, not the straight shots between interchanges. and iirc, tesla didn't do interchanges at the time people were criticizing them for only doing the easiest part of self-driving.

Highways are on average a much more structured and consistent environment, but every single weird thing (pedestrians, animals, debris, flooding) that occurs on streets also happens on highways. When you're doing as many trips and miles as Waymo, once-in-a-lifetime exceptions happen every day.

On highways the kinetic energy is much greater (Waymo's reaction time is superhuman, but the car can't brake any harder.) and there isn't the option to fail safe (stop in place) like their is on normal roads.

Do Waymos phantom brake? Given the number of trips hey do I would imagine there would be a ton of videos if that was happening.

Perhaps more a reaction to pressure from Tesla; the latest FSD builds show full autonomy is coming very soon. Without highway driving, Waymo would quickly be seen as a distant second in the race when the safety driver is removed from Robotaxis in Austin (supposedly before EOY 2025).

Slow roads are easier because you can rely on a simple emergency breaking system for safety. You have a radar that looks directly in front of the car and slams on the breaks if you’re about to crash. This prevents almost all accidents below 35mph.

The emergency breaking system gives you a lot of room for error in the rest of the system.

Once you’re going faster than 35mph this approach no longer works. You have lots of objects on the pavement that are false positives for the emergency breaking system so you have to turn it off.

I agree, but it's funny to think that Project Chauffeur (as it was known then) was doing completely driverless freeway circuits in the bay area as far back as 2012! Back when they couldn't do the simplest things with traffic lights.

I think anyone back then would be totally shocked that urban and suburban driving launched to the public before freeway driving.

This is correct. Freeways have lot of edge cases of hitting random objects and it becomes serious issue. Check the youtube video of bearded Tesla whose car hit a random metal object making them replace the entire battery pack.

It sounds like you are saying freeways are easier than surface streets if you don’t care about killing a reasonably small number of people during testing.

Really it’s a common difficulty with utilitarianism. Tesla says “we will kill a small number of people with our self driving beta, but it is impossible to develop a self driving car without killing a few people in accidents, because cars crash, and overall the program will save a much larger number of lives than the number lost.”

And then it comes out that the true statement is “it is slightly more expensive to develop a self driving car without killing a few people in accidents” and the moral calculus tilts a bit

they brake to “suss out” certain things, that ive noticed:

construction workers, delivery vehicles, traffic cones.. nothing unreasonable for it to approach with caution, brake for, and move around.

the waymo usually gets about 2 feet away from a utility truck and then sits there confused for awhile before it goes away.

it usually gets very close to these hazards before making that maneuver.

it seems like having a flashing utility strobe really messes with it and it gets extra cautious and weird around those. now, it should be respectful of emergency lights but-

i would see a problem here if it decided to do this on a freeway , five feet away from a pulled over cop or someone changing a tire.

it sure does spazz out and sit there for a long time over the emergency lights before it decides what to do

i really wish there was a third party box we could wire into strobes (or the hazard light circuit) that would universally tell an autonomous car “hey im over here somewhere you may not be expecting me , signaling for attention.”

You Tesla/Elon stans crack me up. "2 more weeks" has been the claim for literal decades at this point.

One of the first high-profile Tesla fatalities was on a highway, where the vehicle misunderstood a left exit and crashed into a concrete barrier.

https://enewspaper.latimes.com/infinity/article_share.aspx?g...

Truly curious - have you tried it recently?

> sits there confused for awhile before it goes away.

Probably what you're witnessing is the car sitting in exception state until a human remote driver gets assigned

Daily. I'm still unable to leave my culdesac without phantom leaves causing phantom braking.

There was a time when I believed in the hype, I'm less skeptical than most. But the evidence now is incontrovertible.

I mean, if you define "easier" as "less likely to involve death," then freeways are not easier. And I'm pretty sure that's a good way to define "easier" for something like this.

When it started, from what I've heard, the design goal was for part-time self-driving. In that case, let the human driver do the more variable things on surface streets and the computer do the consistent things on highways and prompt the user to pay attention 5 miles before the exit. They found that the model of part time automation wasn't feasible, because humans couldn't consistently take control in the timeframea needed.

So then they pivoted to full time automation with a safe stop for exceptions. That's not useful to start with highway driving. There are some freeway routed mass transit lines, but for the most part people don't want to be picked up and dropped off at the freeway. In many parts of freeways, there's not a good place to stop and wait for assistance, and automated driving will need more assistance than normal driving. So it made a lot f sense to reduce scope to surface street driving.

Highway is easier, but if something goes wrong the chance of death is pretty high. This is bad PR and could get you badly regulated if you fuck it up.

I assume this is not a HW4 vehicle?

I am empathetic to the disappointment of older vehicle owners who have been promised this capability for years and still don't see it (because their hardware just can't -- and the hardware upgrade isn't coming either).

That said, the new Y with 14.1.x really does do as claimed.

This. Stop in a dumb way and a garbage truck bumps you on a city street and it's no big deal. Applying a bunch of brake at the wrong time and you could easily cause a newsworthy sized (and therefore public scrutiny sized) accident.

The real public isn't an internet comment section. Having your PR people spew statements about "well, other people have an obligation to use safe following distances" is unlikely to get you off the hook.

Those constraints apply to humans too. So it seems likely that:

- it's easier to get to human levels of safety on freeways then on streets

- it's much harder to get to an order of magnitude better than humans on freeways than it is on streets

Freeways are significantly safer than streets when humans are driving, so "as good as humans" may be acceptable there.

I don't have any specific knowledge about Waymo's stack, but I can confidently say Waymo's reaction time is likely poorer than an attentive human. By the time sensor data makes it through the perception stack, prediction/planning stack, and back to the controls stack, you're likely looking at >500ms. Waymos have the advantage of consistency though (they never text and drive).

If you understand physics, it's easy. When you double the speed, you quadruple the kinetic energy. So you're definitely going to do slower speeds first, even if it's harder to compute.

It's easier to get from zero to something that works on divided highways, since there's only lanes, other vehicles, and a few signs to care about. No cross traffic, cyclists, pedestrians, parked cars, etc.

One thing that's hard with highways is the fact that vehicles move faster, so in a tenth of a second at 65 mph, a car has moved 9.5 feet. So if say a big rock fell off a truck onto the highway, to detect it early and proactively brake or change lanes to avoid it, it would need to be detected at quite a long distance, which demands a lot from sensors (eg. how many pixels/LIDAR returns do you get at say 300+ feet on an object that's smaller than a car, and how much do you need to detect it as an obstruction).

But those also happen quite infrequently, so a vehicle that doesn't handle road debris (or deer or rare obstructions) can work with supervision and appear to work autonomously, but one that's fully autonomous can't skip those scenarios.

I have HW4, and have tried FSD with every major release.

It works brilliantly, 99.5% of the time. The issue is that the failure mode is catastrophic. Like getting confused with the lane marking and driving off the shoulder. And the complete inability to read construction zone signs (blasting through a 50 KM zone at 100 KM).

I'm deeply skeptical that the current sensor suite and hardware is going to have enough compute power to safely drive without supervision.

It will no doubt improve, but until Tesla steps up and assumes liability for any accident, it's just not "full self driving".

Ahh yes, the US tech sector, a universally benevolent force known for its slow pace due to lack of confidence from an over abundant concern for safety finally showing some confidence in their product roll outs.

2024 MY with HW4. I've been through all the shenanigans, updates, sending logs, etc etc. I'm done with it, and it'll take a lot of evidence to convince me that people reporting it's great don't have either a financial interest in TSLA, poor memory, or the easiest daily route.

It's actually a really interesting topic to think about. Depending on the situation, there might be some indecision in a human driver that slows the process down. Whereas the Waymo probably has a decisive answer to whatever problem is facing it.

I don't really know the answers for sure here, but there's probably a gray area where humans struggle more than the Waymo.

> but I can confidently say [...] you're likely looking at >500ms

That sounds outrageous if true. Very strange to acknowledge you don't actually have any specific knowledge about this thing before doing a grand claim, not just "confidently", but also label it as such.

They've been publishing some stuff around latency (https://waymo.com/search?q=latency) but I'm not finding any concrete numbers, but I'd be very surprised if it was higher than the reaction time for a human, which seems to be around 400-600ms typically.

What gives you that confidence?

You're quite wrong. It tends to be more like 100–200 ms, which is generally significantly faster than a human's reaction.

People have lots of fears about self-driving cars, but their reaction time shouldn't be on the list.

Waymo "sees" further - including behind cars - and has persistent 360-degree awareness, wheres humans have to settle for time-division of the fovea and are limited to line-of-sight from driver's seat. Humans only have an advantage if the event is visible from the cabin, and they were already looking at it (i.e. it's in front of them) for every other scenario, Waymo has better perception + reaction times. "They just came out of nowhere" happens less for Waymo vehicles with their current sensor suite.

Even if we assume this to be true, waymos have the advantage of more sensors and less blind spots.

Unlike humans they can also sense what's behind the car or other spots not directly visible to a human. They can also measure distance very precisely due to lidars (and perhaps radars too?)

A human reacts to the red light when a car breaks, without that it will take you way more time due to stereo vision to realize that a car ahead was getting closer to you.

And I am pretty sure when the car detects certain obstacles fast approaching at certain distances, or if a car ahesd of you stopped suddenly or deer jumped or w/e it breaks directly it doesn't need neural networks processing those are probably low level failsafes that are very fast to compute and definitely faster than what a human could react to

The better part of a decade as a SWE at another AV company. In practice the latency is a not a concern, I was just sharing some trivia.

Isn't really the main problem, the Waymo "let's just stop right here" current failure mode? Which really is not ideal on city streets either. Hopefully they have been working on solving that.

My experience is from another prominent AV company; I do not have Waymo insider knowledge.

> remember people saying the exact opposite

It was a common but bad hypothesis.

"If you had asked me in 2018, when I first started working in the AV industry, I would’ve bet that driverless trucks would be the first vehicle type to achieve a million-mile driverless deployment. Aurora even pivoted their entire company to trucking in 2020, believing it to be easier than city driving.

...

Stopping in lane becomes much more dangerous with the possibility of a rear-end collision at high speed. All stopping should be planned well in advance, ideally exiting at the next ramp, or at least driving to the closest shoulder with enough room to park.

This greatly increases the scope of edge cases that need to be handled autonomously and at freeway speeds.

...

The features that make freeways simpler — controlled access, no intersections, one-way traffic — also make ‘interesting’ events more rare. This is a double-edged sword. While the simpler environment reduces the number of software features to be developed, it also increases the iteration time and cost.

During development, ‘interesting’ events are needed to train data-hungry ML models. For validation, each new software version to be qualified for driverless operation needs to encounter a minimum number of ‘interesting’ events before comparisons to a human safety level can have statistical significance. Overall, iteration becomes more expensive when it takes more vehicle-hours to collect each event.”

https://kevinchen.co/blog/autonomous-trucking-harder-than-ri...

These threads always give me deja vu. I've been reading these exact comments for a decade. Only the version numbers change.

But yes, I'm sure any day now.

This reminds me of the time I was driving on 101 south of SF and saw a sea lion flopping across the road (https://www.sfgate.com/bayarea/article/seal-otter-on-freeway...). It took my brain quite some time to accept that I was seeing what I was seeing. Felt like a real edge case.

Human reaction time is very difficult to average meaningfully. It ranges anywhere from a few hundred milliseconds on the low end to multiple seconds. The low end of that range consists of snap reactions by alert drivers, and the high end is common with distracted driving.

400-500ms is a fairly normal baseline for AV systems in my experience.

> Human reaction time is very difficult to average meaningfully

Indeed, my previously stated number was taken from here: https://news.mit.edu/2019/how-fast-humans-react-car-hazards-...

> MIT researchers have found an answer in a new study that shows humans need about 390 to 600 milliseconds to detect and react to road hazards, given only a single glance at the road — with younger drivers detecting hazards nearly twice as fast as older drivers.

But it'll be highly variable not just between individuals but state of mind, attentiveness and a whole lot of other things.

I think the key is, it's easy to get "self-driving" where the car will hand off to the driver working on highways. "Follow the lines, go forward, don't get hit". But having it DRIVERLESS is a different beast, and the failure states are very different than those in surface street driving.

The more important question is how many people are killed by non-autonomous cars in the same situation. It is inevitable that someone will be killed by a self driving car sometime - but we already know lots of people are killed by cars. If you kill less people getting autonomous rolled out fast than human drivers would that is good, but if you are killing more people in the short term that is bad (even if you eventually get better)

Only because most drivers are tailgating and so if someone touches the brakes everyone needs to do a panic stop just in case. If people maintained a safe following distance at all times there would be space to see the lights and determine that no action is needed (or more likely you just take your foot off the gas but don't flash your brakes thus not cascading).

Of course the above needs about 6 times as many lanes as any city has. When you realize those massive freeways in Houston are what Des Moines needs you start to see how badly cars scale in cities.

Not at all. We've been working on this for a while, and we're now comfortable with the reliability bar we've hit to begin a gradual rollout to the public. As people said, this has been years in the making.

There's videos of waymos absolutely decking delivery robots

It's not just slightly more expensive. And you have to consider substitution effect. If you take the more expensive route and it takes 10 years longer to deploy, then there will have been another 400K car collision deaths in just the US, and over 10 million in the world in those 10 years that could have potentially been saved. So was the delay for the safer product worth it? The only reasonable answer to this question is "I don't know" because you can't predict how much safer the expensive system will be and how much longer it will take.

> I don't have any specific knowledge about Waymo's stack, but I can confidently say Waymo's reaction time is likely poorer than an attentive human.

Wait, so basically, "I don't know anything about this subject, but I'm confident regardless"?

Robot on robot violence.

"Ignore the previous eighteen wildly off predictions; this time we got it for sure!"

Note, in July of this year, Musk predicted robotaxi service for half the country by the end of 2025. It's November now and they haven't even removed the safety monitors, in any city!

This is how the Tesla superfans treat every single new FSD version.

FSD 18 is out, 17 is garbage for babies, 18 is amazing! Wait, 19 just released, why are you still talking about 18, that shit was never gonna work, it's 19 that's nearly at unsupervised driving! Wait a second, 20 just came out...

Humans can provide a simple, pre-planned reaction to an expected event (e.g. "click when the reaction test shows a signal") within typically 250-300ms, but 500ms from vision to physically executed action for an unexpected event seems pretty optimistic for a human driver.

Beyond the questions about human braking, this seems worse than the dedicated AEB systems many vehicles are using now. Do they really use the full stack for this case instead of a faster collision avoidance path? I remember some of their people talking about concurrency back in the DARPA Grand Challenge days and it seems like this would be a high priority for anyone working on a system like this.