The 5th-generation Waymo Driver | Hacker News Reader

The 5th-generation Waymo Driver

blog.waymo.com

214 pointsthrowaway3157155 comments

I look forward to the first test of 1000+ self driving vehicles going through a swerving road, multiple merge, plus toll plaza situation (like merging onto the lower deck of the GW bridge east-bound). If this technology is as good as claimed, it really should put human drivers to shame. Imagine no more traffic jams because of the needless “jockeying for position”, or the drivers who are clueless about the boundaries of their own vehicle. As someone who has been driving for over 30 years, I honestly can’t wait for the day this promise materializes.

Of course, if self driving cars defer (in a predictable manner) in order to prevent collisions, I can easily see how one or two human drivers could reap havoc to their advantage. It would become a new sport.

Will look forward to it too, if we'll ever see it.

Traffic control if a big part of the problem, if not a half of it.

If you had 100% pedestrian free roads, with ideal road surface, and radio signalling, where a car becomes indistinguishable from a train for control purposes, then even with current state of computer science it becomes quite real.

This is the route most Chinese companies are progressing on. They are not making a self-driving car, but a "self-driving road." Something akin to air traffic control system for cars.

And this is only for things like public transport, garbage dumpsters, cleaning, and other public utility vehicles.

The bus gets a radionavigation receiver, radar for collision avoidance backup, and receives commends over the air.

No road is 100% pedestrian free. When cars break down on the road the occupants tend to become involuntary pedestrians. Wild animals also exist.

In other words, railroads, just without the need for physical rails. Great, cool, useful. Not quite the usual expected result, though.

I can see where it would be valuable, but it may cause more traffic (and therefore traffic jams). If it is cheaper and easier to drive and people can live further away from work with no penalty, it will lead to lots more cars on the road. It will lead to more sprawl, less walking and biking so more obesity, and massive damage to the environment.

Increased road usage taxes.

"For your own good."

It should be easy to gather political consensus for laws to prevent what is metaphorically like pulling a fire alarm for fun. Society would similarly disapprove of seizing up trains by threatening to jump.

All this to avoid building new public transit infrastructure. Crazy. Many of the thorniest transit problems in dense city centres are better solved by improving mass transit options, not giving even more right of way to cars.

Mass transit will always be worse than driving, unless traffic and parking are both really bad and transit can skip the traffic somehow. This is because of the last-mile (and first-mile) problem and the fact it can't be truly on-demand. The only way to solve these issues is with some system that starts and ends in a personalized trip, which would ironically probably be more restricted (it would probably have rails and fences)

Also, most mass transit have similarly restricted areas to roads, for example you can't walk on the light rail in seattle any more than you can walk on the road it's built next to.

Idk Chicago's transit is really good. It's getting crowded in some places, but still great.

It of course has an advantage - it's been around so long that the city has grown around major transit lines, so the things you want to do are all grouped around & easily accessibly by transit.

If you take a city like LA or Houston that has zero centralization and just slapped transit lines down, they would have trouble servicing everything and average person wants to do - maybe you can get to work on the line, but not the gym or grocery store.

Still, it would reduce car traffic, and over time it would encourage the desired grouping around transit lines that is part of what makes the transit so helpful in the first place.

I don't think it could ever fully replace cars in most places in America, as they are so spread out and many people will still want to travel between cities by car. However, even just getting most people to get to their 9-5 jobs via transit instead of car would be a massive reduction in congestion & pollution.

Living in a city that has grown around 100+ years of rail infrastructure, mass transit is slightly better than driving (1 hr MT x 1.5 hr+parking car). Sure, there's some walking, but with at most 30-minute tick on suburban rail and under 3 minute tick on city rail, a little planning goes a long way. (Also, car gets about 5x the cost per km, even before the public transit long-term ticket subsidies apply)

I think in many cases zoning and the way land taxes are done are a big problem holding cities back. It's not legal in many cities to make things walkable, bikeable, transit friendly due to laws like single use zoning and parking minimums.

GW bridge: George Washington Bridge between New York City and New Jersey State, United States

I think this shows the tricky bit: https://www.google.com/maps/place/George+Washington+Bridge/@...

Edit: of course it might be the other side... https://www.google.com/maps/place/George+Washington+Bridge/@...

This is a newbie question, but why is the lidar/radar necessary? Is it solely because computers aren't yet able to extract as much information from video as humans can from sight? Theoretically, could self driving cars use just video with enough compute resources and the right ML algorithms?

There are a lot of people here very confidently stating that LIDAR isn't necessary but seemingly forgetting to link to any actual evidence of fully autonomous vehicles that work without it. The issue is that human vision isn't just vision, it's vision linked with a ridiculous general intelligence that we're no where close to reproducing. So humans are great at categorising the objects they see, and reasoning and how they'll behave, which is why we don't generally need LIDAR.

So it's just as much about how humans can reason from sight as it is about how good their sight actually is. It's very possible that by the time you've built a computer capable of the same sort of reasoning that humans do, you've actually solved, well, artifical general intelligence and you're no longer designing a car at all - because the AGI took that job ages ago.

We,humans,are very good at some things that will probably take decades for computers to get anywhere near such a level. For instance, you driving a car and there's a huge truck ahead of you transporting big logs. There isn't much traffic and eventually you get close enough to the truck to see that the structure holding the logs is somewhat loose.Your brain can instantly create a scenario of what would happen if the logs would start falling on the road.This alone gives you enough info to act upon.What you do( slow down or try to take over and warn the driver at the same time) is different question.In probably less than a second your brain recognised the type of cargo the truck had, it's physical properties, some outcomes should parts of the truck fail and so on. For us it's easy,for computer this would probably take days worth computation.

Not just humans, bit even the tiniest of insects are capable of navigating with ease in complex environments at a speed and adaptability we are not even close to match. I'm pretty sure that having the raw intelligence of an ant brain in a mars rover would expand our exploration capabilities by several order of magnitudes.

Nah, insects just have very high body durability at the speeds they move.

>There are a lot of people here very confidently stating that LIDAR isn't necessary but seemingly forgetting to link to any actual evidence of fully autonomous vehicles that work without it.

Funny, I was not aware of any fully autonomous vehicle that actually works with or without it.

I'm not sure that human uses that general intelligence when driving and when decisions must be made quickly. I think that brain uses shortcuts and does not really reason about things. And those shortcuts are likely work similar to trained neural networks. That's one reason why novice drivers have problems: they don't have those shortcuts in their brains yet, so they have to reason and that takes valuable time and attention.

Though I think that robotic automobiles must perform better than humans, so equipping them with better sensors makes perfect sense, regardless of algorithms.

I think it's always best to avoid comparing neural nets with biological neurones and their networks. The comparison just doesn't hold since they are completely different in almost every way.

Also I think that it actually proves that humans use their general intelligence to drive, since Humans can learn on the go and add almost any type of new knowledge and act upon it without even realizing it. The brain is not looking for anything in particular, just any type of information that would be relevant to driving.

According to your average SV pitch deck, we are 6-10 months away from reproducing general intelligence. Add in a sprinkling of Pascal's Wager, and your funding is almost guaranteed.

Which goes back all the way to the Dartmouth workshop, which hoped to get most of the work done in 6-8 weeks. That was in 1956. Is most of the work done? Well there you go. (Your hint that it's a good way to get paid for such work holds, of course, but the expectation that Problems will get Solved...unlikely)

There's a whole lot of challenging information that is completely natural and intuitive for a human to understand but fiendishly difficult for a ML algorithm to figure out. There's some cues that I'm sure we probably won't be able to use until we create a genuine artificial general intelligence. If you're driving and you see someone standing at a crosswalk it's intuitive just by looking at them if they're waiting for you to pass, about to walk out into the road, panhandling, etc. You can put yourself in their shoes and make a reasonable prediction as to what they are thinking about doing. On a previous HN thread there was a commenter that had a run in with a Waymo car while riding a bike. He was coming up to a 4 way stop, and yielding to the Waymo car. If he was balancing on his pedals, still stationary, the Waymo car would stop and not proceed through the intersection, apparently interpreting his pose as if he's about to proceed through the intersection itself. A human driver wouldn't have an issue with that, but you can imagine the training data would show that when a bicycle is stopped, the rider puts a foot on the ground.

Generous helpings of lidar and radar to augment cameras is a crutch to help compensate for the lack of 500 million years of unsupervised learning that went into our visual cortex.

It might look intuitive just by looking at them, but good drivers won't trust that intuition, because humans are very unpredictable species. So good drivers will slow down to have enough time to react if necessary. You should not trust "reasonable prediction" when you're talking about death danger.

Which means "do not drive, ever." The actual risk appetite for (human) driving is very different from the advertised one; the heated discussions on SDV just bring this to light.

Lidar or radar won't help you with any of those situations though

For those specific cases no, but it makes a great source to identify physical objects and their placement in the world. Obviously that can be done to a sufficient degree with just stereoscopic vision, but look at all of the autopilot fatalities from Tesla. All too often vision and radar data gets misinterpreted as "That must be an object to the side of the road" right up until it crashes into it. With lidar you can be sure you're not just looking at a radar reflection or getting the perspective wrong on a camera. Almost all of the large objects that you want to make sure you never hit will show up well with lidar.

https://en.wikipedia.org/wiki/Tesla_Autopilot#Incidents

Just about every one of those fatalities can be summed up as "Tesla ignores large stationary object directly ahead". Lidar would have detected all of those objects and most likely prevented every one of those accidents. I think Tesla currently has the best vision and radar only system out there, so either the state of the art doesn't quite cut it yet without lidar, or there's a ML engineer at Tesla that really really hates fire trucks.

Or Telsa doesn't actually provide Full Self Driving yet and people shouldn't be watching their phones.

It is noticing stationary objects, because sometimes it breaks when the car approaches a bridge going overhead, which is also bad when the car too close behind doesn't.

You have to ignore some objects in front of you (even ones heading directly towards you) because you're going round corners, so it's never cut and dried

It isn't necessary. As you suggest, humans have only normal eyes, not laser eyes, and we still manage to drive.

Yes and no. It's also just nice to have more different kinds of data available. Just because humans don't have laser eyes doesn't mean we can't try to do better than that.

Depth from stereo is not ML, but yes.

> Depth from stereo is not ML, but yes.

Tesla in particular, and others are moving towards getting depth from ML. Yes, you can do dumb coincidence-finding, but there's a lot of corner cases (leafy objects, specular reflections, etc) that screw this up. Humans don't just use coincidence finding, but use all kinds of other clues (size, texture gradients, monocular parallax, shadows, linear perspective, attenuation from haze, etc) to infer depth.

> Humans don't just use coincidence finding

Maybe not just, but we do use it, except we do it by converging our eyes for _actual_ coincidence within a small focal center instead of ambiguous coincidence everywhere in the field of view like CV stereo typically tries to do. Gimbaled cameras that (metaphorically speaking, of course) "look" where drivers are supposed to look with an attention model instead of being statically coplanar could do this too.

> except we do it by converging our eyes for _actual_ coincidence within our focal center instead of ambiguous coincidence everywhere in the field of view.

Only at very close range. That's convergence, and mostly a 5m and less phenomenon (not so relevant at driving ranges).

Tesla is using data from the radar to annotate the images from the cameras with distance before running it through the ML training, but for driving it is using a combination of all (stereo vision, ML and radar) to produce the final number.

The human neural network is of course fine tuned to this and my father who is blind on one eye will actually move his head sideways a bit to judge distance when he is driving :)

Yah. Waymo is doing the same with the lidar, which provides both higher accuracy and higher resolution "ground truth" distance data. And yes, everyone uses sensor fusion and filtering, not just "ML".

> we still manage to drive

not amazingly well, though, because there are millions of victims each year

SDCs are supposed to be better than us, so more sensors make sense.

I'd argue that the majority of victims are the result of humans suppressing their judgment. E.g. not because a human was incapable to drive safely, but purposedly chose to drive unsafely: overtake where it is limited visibility, speeding when conditions do not allow for it.

Attention-related crashes are likely a good chunk as well.

Also unsafe lefts and ran/stale lights. That's probably most all of them between our comments.

It isn't "necessary" for small values of "necessary", ie if you have a human brain hooked up to human eyes. It is extremely necessary if you only have a computer and some primitive software, like we do now, as in "self-driving won't work as well otherwise".

Yet depth from stereo is, to some extent, AI because the matching of pixels that corresponds to the same physical point is not trivial and uses heuristics.

Depth from stereo gets worse at distance, while LIDAR and RADAR read depth directly.

Even in broad day light and hith High Res videos and with ML algos, we cannot extract all information accurately that are needed. For example, at what velocity the other vehicle is moving and also at what distance we have an obstacle etc. is very difficult to extract from camera alone and impossible when it is night or extreme weather conditions. Hence we need Radar & Lidars also for ADAS & Self driving applications.

Radar: It can accurately measure Distance and Velocity information of objects around ego vehicle and also can track objects. It works well in all weather conditions (Day/night/rain/fog etc). Lidar: Good distance measurement, Rich in data (3D Point cloud) for ML, OK in doing classification (pedestrain, bicyclist etc) even in nights. But expensive sensor.

Besides the other reasons it's also that the dynamic range of eyes are way higher than the cameras they can use here. And humans have access to mechanisms that allow them to handle washouts.

They're just building in the equivalent of putting in your sunglasses or looking off to the side when the sun is in your eyes or readjusting your seat position so it isn't hitting your eyes.

There's no practical reason to limit autonomous vehicles to human-visible signals. The point is to make a car that drives safer than a human. Since we can't make a car that's smarter than a human, we can at least even the odds and make it see much better.

Actually the reason is the most practical one of all; money.

A system which costs too much to widely deploy and maintain will never see widespread deployment.

> This is a newbie question, but why is the lidar/radar necessary?

It's not, in theory. It is, in practice.

> > Is it solely because computers aren't yet able to extract as much information from video as humans can from sight?

Yes. Currently, you cannot, for any amount of money, buy a camera that has equivalent visual acuity to the human eye.

I’m not convinced that the eye is conclusively better than any camera out there. The human brain is incredible at stitching together visual input into a coherent view of the world. The retina has high resolution but covers an incredibly small field of view; rapid eye movements make it possible to fake a larger effective field of view. And I’m not convinced that the retina’s resolution is actually better than that of high-end camera systems. The eye’s performance in the periphery is decidedly worse than that of a wide-angle camera.

Dynamic range is a challenge for cameras, but we have high-dynamic-range imaging nowadays (both in software, i.e. exposure combining, and hardware). I don’t think the eye is significantly superior here.

Low-light used to be a strength of the human visual system, but I think modern computational imaging systems have caught up (plus, human night vision was really never that good compared with other animals).

So in short: I dispute the idea that cameras don’t have the acuity of the human visual system, nowadays. I’d like to know in which aspects you believe human eyes to still be superior, from an optical point of view. Obviously - the human brain and visual cortex is something that computers are nowhere close to.

> So in short: I dispute the idea that cameras don’t have the acuity of the human visual system, nowadays. I’d like to know in which aspects you believe human eyes to still be superior, from an optical point of view.

Dynamic range. Your eyes can pick up subtle details in a scene with very bright lights, and very dark shadows.

No video camera currently exists that can take a good video from inside a city, at night, with a starry night sky. Either the stars, or the lights, or the shadows are going to look like crap. Your eyes can trivially handle such a problem.

our vision system is mostly lousy optics coupled with outstanding processing.

The ability to capture both stars and a daylit scene using the same sensor is pretty hardcore. But yes, there's some fancy processing layer which is far harder to replicate than the optics. Last time I tried if I point a generic (not specialised for the task) camera at the night sky I get blackness but my eyes see stars.

On the optics, I'd say the human eye is specialized, but not on the whole outstanding. Fantastic dynamic range, lousy spatial and time resolution, a very good tracking mount. It's just compromised on the things (resolution) that can be filled in with fast scanning over a scene + post-processing.

This seems nonsensical given the telescopes we use to see things that cannot be seen with the naked eye.

Open the lid on that telescope during the day, and you'll fry the telescope.

Your eyes can operate - and operate well - in an incredibly broad array of lighting conditions. No single camera exists that can currently do that.

Computer vision is a very noisy application. Unless you have some extremely good software filters, that noise will reduce reliability, what in a car may end in disaster. Even humans are fooled by it once in a while.

And even if Google has those extremely good filters, there is an entire different order of magnitude of testing until they can be sure. They are probably jut trying to get something out of the lab by using reliable sensors, and letting optimizations for later.

Why should we pursue biomimicry to that extent? My go to example is to compare aircraft to birds, so I could rephrase your question as why are the engine/fixed wings necessary? Is it solely because material engineering hasn't advanced yet to enable airplanes to flap their wings to stay airborne like birds?

Human eyes and brains are completely different from solid state electronics with different constraints and advantages.

There is in fact a pretty long history of biomimicry in aviation though. I've met many aerospace engineers (including a former NASA chief scientist) who would take your question very seriously. The answer may well be "no, that isn't the only reason", but it's not trivial to arrive at that answer.

Self-driving tech is still at an immature stage while the bar set by public is super high; even human-level self driving might be unacceptable for pervasive adoption. Anyway you can use both Lidar and camera. It's not too late to worry about this cost-performance trade-off after making it actually work.

It isn't necessary and the best example of this is Tesla's autopilot. It uses no Lidar component and Elon Musk's bet is that self-driving cars can just use just cameras, basic radar and ultrasonic sensors with enough compute resources and the right ML algorithms to perform better than a human.

He's been a long time anti lidar proponent because of the costs involved and the aesthetics. He's also betting that the amount of data Tesla receives from its customers, and the neural net they have can achieve autonomous driving with its current hardware stack.

“In my view, it’s a crutch that will drive companies to a local maximum that they will find very hard to get out of,” Musk said. He added, “Perhaps I am wrong, and I will look like a fool. But I am quite certain that I am not.

"Despite being a fancy and expensive technology, LiDAR provides surprisingly little advantage over a combination of cameras and radar. Radar, for example, is much better in the rain and other limited visibility scenarios, because it is based on radio waves rather than light waves. Radio can penetrate through some objects and bounce back from others, thereby “seeing” the environment along a different dimension."

an excerpt from a quora answer on why the Tesla stack could be better than Lidar: https://www.quora.com/Why-dont-Tesla-cars-use-Lidar-like-mos...

Here's a video of Tesla's autopilot perceives its environment : https://www.youtube.com/watch?v=fKXztwtXaGo and here's a video of how Waymo perceives its environment: https://www.youtube.com/watch?v=OopTOjnD3qY

The question is if the Lidar adds incremental value or exponential value, and I think it's just incremental by looking at those videos.

Tesla is not a good piece of evidence. They registered a grand total of 12.2 autonomous miles in 2019. Their "Autopilot" is a particularly fancy driver assist system; if you take their marketing at face value and treat it like an autonomous driving system you are putting yourself and everyone on the road at risk.

That's fair, I've just taken the marketing at face value.

Though I ended the answer at the end asking whether Lidar adds incremental or exponential value? Do you think it adds exponential value ?

I'm not an autonomous car engineer so don't understand the nuances but from whatever basic information I've read it doesn't seem like Lidar's add exponential value.

edit: Just to add, 5 million Waymo miles have been driven with a driver that controls the car and Waymo has had an accident too - https://www.wired.com/story/waymo-crash-self-driving-google-...

Also Waymo has way lesser cars on road than Tesla

Exponential. It gives range and shape data, which a pure-optical system needs to infer from a 2D image. This kind of image processing is still an open problem in ML.

The usual metric for self-driving car success is "disengagements per mile", ie how frequently a driver needs to intervene to avoid a crash. From my anecdotal readings of Tesla Autopilot reviews, it's on the order of 0.1 per mile. For Waymo and Cruise, it's on the order of 0.01 per THOUSAND miles. That's a very different definition of "driver" than the one that Tesla Autopilot requires.

I don't know the total number of miles on all Teslas on Autopilot, but it has had much more than one accident.

EDIT: and that Waymo crash was not a self-driving error; it was T-boned by a human-driven car running a red light.

Fair, depth perception from 2D isn’t there yet through ML, but mixed with radar can it be effective enough.

The reason I assumed it works is that lidar on the article above seems more like a redundancy. Because their camera system have the short range covered and radar has the long range covered. Lidar seems to augment over it.

Though the order of disengagement is a great stat, that definitely shows how much better waymo is compared to Tesla

> but mixed with radar can it be effective enough.

The usual solution is actually lidar + optical; lidar gives much better spatial resolution than radar, which is why it's been the standard going back to the DARPA challenge. You really want to have good spatial resolution in order to distinguish e.g. bikers and tail-lights and road signs for your optical systems, which radar typically isn't good enough for; that's the point of that qualifier in "imaging radar". Still probably worse performance (i.e. time and spatial resolution) than lidar, but better range and weather resistance.

(The previous generation of Waymo cars already had one lidar on top; the radar and the close-range lidars are the new additions.)

Their AutoPilot has driven billions of miles. Yes, the driver must still be paying attention and ready to take over. That doesn’t mean the system wasn’t driving.

Miles per disengagement I’m sure is not too high. That would be a good metric to have. But total miles is still ~2 billion.

> He's been a long time anti lidar proponent because of the costs involved and the aesthetics.

I can't help thinking that, whatever the merits of Lidar, Musk is boxed in, because Tesla has sold hundreds (tens?) of thousands of "self driving packages" for cars not equipped with Lidar, so changing course would not just mean raising prices on new cars, but retrofitting large numbers of existing cars at a ruinous cost.

That’s a fair point. He does have a bias to sell that story.

> Elon Musk's bet is that self-driving cars can just use just cameras, basic radar and ultrasonic sensors with enough compute resources and the right ML algorithms to perform better than a human.

I'm not sure that unsubstantiated claims from Elon Musk are actual evidence that lidar isn't necessary.

It's substantiated by the fact that humans don't have radar and ultrasonics, just two cameras on a swivel and a lot of signal processing, and they succeed at operating a car to five-nines reliability measured in miles traversed. So Musk's bet isn't completely bonkers; he knows of at least one reference system that does the task with fewer sensors than even shipped with the Tesla.

... but we do want the SDC to do better, and there are failure modes that human perception is also vulnerable to generally. In addition to closing the gap faster on solving the problem without a copy of the human perception wetware, the LIDAR signal might also improve on those perception error states and be worth keeping in the design even if it could be done with cameras alone (or camera + radar + ultrasonic).

I mean, sure, humans don't have radar or ultrasonics.

However, computers don't have human brains, and AI doesn't provide _anything like them_.

Defininitely agree, and I think that's the devil hiding in the details about Musk's bet that is worth surfacing: he's making the bet "We can just build a computer as good at this complex highly-variable task as a human being," and it's a bet people have been making and losing for decades.

Some day, someone will make that bet and be right. I haven't put my money on this team and this project. ;)

> the Tesla stack could be better than Lidar

From the systems I've worked with it's usually AND and not OR, you use both a Lidar and a Radar. The Radar images I've seen were quite lousy and are not 100% interference prone.

The problem is Tesla autopilot only works on highways and has been implicated in a number of crashes. Waymo seems to be taking the much more cautious approach of using every advantage they can get. Perhaps one day there will be self driving cars without lidar. But for now I think Waymo's results speak for themselves.

A few people died following that bet though

I don't have a Tesla nor am I a fanboy. Just talking about the tech.

This is biased information as there are way more Tesla's out there compared to Waymo's.

Waymo has had accidents too https://www.wired.com/story/waymo-crash-self-driving-google-...

Your evidence is a waymo car getting t-boned at an intersection?

By a car running a red light, no less.

I think it’s absolutely fair to expect autonomous vehicles to have awareness of a probable side collision in an intersection and be able to speed up or slow down to avoid it.

I often see oncoming cars rushing to make a left turn past when their arrow has turned yellow and red, and so I know not to enter an intersection even though my light has turned green.

Autonomous systems in theory should be better than humans at this because they can track all surrounding objects and trajectories, not just ones they are looking at with one set of eyes.

I think the accident rate is kind of a meaningless stats. You can have a low accident rate by carefully controlling the conditions under which you test. Not many accidents means they aren’t pushing the envelope. That’s probably a good thing on public roads. It’s also why the system is not available for general public use except under extremely controlled routes and close (remote) supervision.

I think it’s great we have (at least) two mega-companies in a race trying different approaches to reach a solution. There are good points for and against both approaches. This is what makes life interesting, you can’t just run the numbers to predict the future.

Thing is, Tesla is not one of the mega-corporations in the running; it's Waymo (Google), Cruise (GM), and possibly Apple.

Linking to an article about a waymo car being hit by a car in a side collision seems pretty disingenuous when comparing to the Tesla accidents that people usually talk about

A camera (or a binocular camera setup) could only ever return depth planes. Lidar returns true 3d forms. Also, depth plane evaluation from cameras is not 100% reliable. Things like reflections, glare and suchlike screw it up.

This is nonsense. There is no difference between the data that binocular cameras return and LIDAR, other than that LIDAR is more reliable and accurate. Your distinction between "depth planes" and "true 3d forms" is gobbledygook.

Maybe I did not express myself accurately enough. I work in VFX, and my language is not always as precise as it could be. But certainly for us this is the difference in application.

Using a two camera input, the best we could hope to get was a depth map. Typically, it is reliable for at most one or two reliable levels of depth, useful for foreground/background separation. Maybe also a crude depth map, useful for fog. With Lidar we could all that plus normal information (i.e. face orientation).

However, I concede your point. The core difference is one of accuracy.

Humans are a pretty poor baseline.

A fully autonomous driving system that was only as good as a human would not gain much traction and the company would be on the hook for some serious legal damages if it became popular.

You make it sound like professional drivers are currently personally liable when they cause accidents. Is this true? (I have no idea)

Lots of debate on the topic (clearly :))

But arguably very little progress (none?) in technology, automation, and industrialization has been through dogmatic replication of biological systems.

Planes don't fly like birds...

Yeah basically LIDAR is better than vision for working out the 3D structure of the world. You can do it from vision, but it's really really hard.

We don't have a working driverless car with nice easy LIDAR data - it would be really silly to try and make one the hard way first, using only vision data (yes I know about Telsa).

It’s all about redundancy. lidars do help though immensely if you integrate it in your ML algorithms currently

Currently it's very difficult to reason about object geometric dimensions / movement from video alone. We (humans) are decent at it and we still make mistakes. "The right ML algorithm" for this is basically a problem as hard as achieving singularity.

Theoretically, since humans drive "ok" using just 2 eyes, you must be able to do it with two cameras and a sufficiently complex brain.

Of course, why would you not avail yourself of more sensory information if you can, provided the cost is not overly burdensome.

Picture this scenario: While looking at a TV screen; what is the difference between a photograph of a car (on the screen) vs a real car (on the screen)?

The only thing a camera can offer is input for pattern recognition. Lidar/Radar offers context.

I'm not suggesting that all you need to fool computer assisted driving is a picture of an empty road taped to the front of the camera; but if the computer can't tell the difference between an optical illusion and reality then I think they need more data inputs.

We don't need computers to be able to handle white-out blizzard driving conditions, but that 'rare' occurrence perfectly illustrates the singular limitation of a camera-based pattern-recognition system.

How do you think a person distinguishes something on a TV screen and the real thing? We don’t have any kind of 3D perception like Lidar.

> With our fifth-generation hardware, we’ve simplified the design and manufacturing process so it can be production-ready, and our latest sensors deliver more performance than ever before, at half the cost of our previous generation.

Has LIDAR just gotten that much cheaper?

It's not uncommon for new prototype-level technologies to become significantly cheaper when you (i) improve processes and (ii) increase scale.

Yup - on the order of halving in price every two years for equivalent models (https://www.thedrive.com/tech/17297/velodyne-just-cut-the-pr...)

They aren’t trying to sell their cars to the general public.

That last photo looks proto-cyberpunk af, straight-up Demolition Man/GitS. They've certainly got the looks down pat.

> will enable the scaled deployment of the Waymo Driver

I've been disappointed by the lack of progress in scaling their commercial service. Hopefully this is what they've been waiting for.

I think this is going to be the big year. Regulatory approval from a bunch of markets and the beginning of fleet deployment, with general availability around most of the US by 2022

Intelligent traffic control is more important right now. Traffic lights with computer vision to know when to change lights and when to clear an intersection with a green arrow.

I wonder if the extra sensors are to stop it accelerating into stationary objects like the Tesla does.

It is my understanding that contrary to tesla's system, waymo's tech is already fully aware of static objects to avoid.

It is also my understanding that none of their cars in computer driven mode drove full speed into a wall or other static obstacle.

The article specifically calls out their radar being able to detect stationary objects, which is unique for automotive radar.

They already had LIDAR, which can easily pick out static objects. The better radar means they have a lower-resolution fallback for poorer weather and longer ranges.

This is adding extra LIDARs, on top of the one they already had. Tesla had its issues because they are trying to do it on the cheap with only cameras, radar, and ultrasonics. That specific failure mode would have been prevented with the hardware that any other self-driving car developer has had for years.

Caveat: I don't live in a developed nation, 3D scanning every traffic light and signs is a futile effort to enable self driving, at least in the next 5 years. 10 years is really optimistic.

All the hardware and software for self-driving has come down in price, but is anyone developing assistive products / super sensing for human drivers?

I for one would really like:

- seeing farther than my eyes can see, warning any sudden speed changes from cars in front

- automated blind spot warning/augmented side mirrors

- seeing beyond corners in tight intersections

Again, business model probably won't be as lucrative as having a fleet of automated cars, but I can see transport companies equipping these in their human-driven fleet to increase safety/reduce fatigue.

Many cars manufactured in the last 2-3 years have the first two features. For example, Honda Sensing: https://automobiles.honda.com/sensing .

Ah, these may not be standard in the developing world but lots of cars have some of these features in the US. Teslas have the ground-bounce radar thing: https://www.youtube.com/watch?v=APnN2mClkmk&feature=youtu.be

Almost every modern car in the US will have the blind-spot warning. My 2018 Subaru does.

There are lots of these. As early as 2018 models of cars you could get features that turn the wheel to keep you in lane, follow the car in front of you at a pace that's safe at a distance you desire, and break in an emergency if the car detects an imminent collision.

yeah it seems like it's almost standard in the US; I can't get those options here, even with a full package of options. I know of automated cruise control and braking, lane keeping systems and such. It's just that manufacturers will always equip as little as they can in a market. US might be helped by regulation/NTHS/insurance. I think there was a fuss about how the same models in Mexico are just a bit cheaper but come with much less safety due to less regulation/pressure.

Your wishlist is already fairly common in most modern cars now, apart from seeing around corners.

A lot of modern cars have some ability to see around corners. This is mostly just used to give a warning when backing out of parking and there is crossing traffic in the lane.

Newer cars do see around corners (relative to the driver), using the rear cameras. When backing out in my newish car, I can see up and down the road in both directions, from the cameras on the rear bumper.

I would like to know more about their imaging radar system. Working in an industry where they are the norm (and quite expensive as a rule) it is interesting to see the technology in consumer oriented solutions.

MM wave radars of non-imaging, low to medium resolution kind are off the shelf parts these days.

True imaging is still a novel application.

What happens when there are many self-driving cars with all these sensors on the road together? Isn't there going to be massive interference?

Nope. The Lidar pulses are encoded uniquely to each sensor, so one sensor can filter out an (already unlikely) direct hit from another lidar. The cameras are IR filtered and also unlikely to get hit by the lidar. Radar is pulse encoded in a similar way to lidar, although possibly more likely to lose performance if surrounded by a lot of other radar units. In general, dirt on your sensor lenses is a lot more annoying than interference from other cars.

Interesting. Do you have any good links to more information about how this works in practice? I guess it's similar to CDMA cell phones.

In a completely autonomous intersection, do you still need stop/go signals, or could all lanes and directions move continuously?

If every vehicle on the road is autonomous, signals are not required. The only exception would be walking crosses but actually the cars should see you and wait for you.

When I say signals, I don't necessarily mean traffic lights.

Would traffic flow continuously from every entry point to the intersection in an "interleaved" manner (setting aside the complexifier that is pedestrian traffic)? Or would there still be a "signal" (visual, over the network, or otherwise) which blocks some vehicles while others pass through? Or, perhaps, the configuration of intersections will change so significantly that this conception of an intersection will be rendered irrelevant?

I saw one driving in SF the other day and was wondering what was up :) usually only see them around mountain view.

5th gen and still beta.

Wow, the tech seems to be progressing nicely. When are they going to commercialize it? They’ve already made steps to make it easier to manufacture(and thus, cheaper), which seems like one of the first off many steps to turn it into a market ready product.

Waymo has the luxury of having a very rich uncle supporting its time to find itself.

There is 0 pressure to monetize Waymo before the technology and business opportunity are at their peak value.

Love how their camera shot is taken from the bike lane.

Is cruise and waymo doing the same thing ? Their cars also look the same.

Probably very similar things - there are only so many useful types of sensors and good places on a car to put them.

Maybe it's time to stop believing big announcements in self-driving cars. I really don't care anymore. You don't need a ton of glossy hype if the product works. But they keep throwing marketing at us.

What big glossy hype? This is a blog post, and not a particularly-high-production-value one. Most self-driving car developers are publicized, if at all, by interested science journalists.

The base model I-PACE has 15-spoke wheels, but they're "gloss sparkle silver". I wonder if it cost more or less to order them in black?

I wonder what their latest hardware suite costs.

I really just want to buy the car with sensors. Give me as the human the full output of sensors and all the objects besides me in a 360 view. That alone would reduce so many accidents.

I think highway driving and stop-and-go slow traffic can be automated significantly. We are very much at that stage.

I love Tesla’s model where humans are in control. Seriously Waymo, I just want to drive your cars and occasionally switch on the “auto” mode on roads that are classified auto-safe in good weather.

Iterate with us.

Would it? In other words, can humans bear a new set of sensors without feeling overloaded? (IDK - at the very least, the peripheral vision cameras could help a lot)

Mostly yes, with a lot of training.

For that, you have to look at aviation. IFR Pilots rely on instruments to fly, these are essentially a "new set of sensors". Using the instruments properly require some intense training. You need to know what to look at and when, it is almost like music, you need to follow a rhythm and do it consciously for a while until it becomes second nature. You also need to resolve conflicts between your gut feelings and the instrument readings: your gut is wrong and the instruments are right, but your brain won't accept that easily.

With planes becoming more and more complex, how to present informations to the pilot is critical. For simple planes you have a set of gauges the pilot need to check periodically. Then it started becoming too much, so a flight engineer was needed to deal with the ever increasing number of gauges. Now there are computer systems synthesizing sensor data to only show the pilot what he needs to fly the plane.

Back to cars, you cannot expect every driver to be trained like an IFR-certified pilot. So showing all sorts of sensor data is going to be counter productive.

That's a great analogy; also, the older a car, the more gauges are on the dashboard: new cars have speed, fuel, engine oil temp, maybe RPM; my oldest car had a battery level indicator and what I think was cooling water temperature. Perhaps this is the way the cars are already going: only show a measure when outside the normal box.

I kinda meant how Tesla shows the 360 view around the car in the dashboard which is fused from different sensors.

I imagine Waymo has even better visual. Seeing all objects, their previous trajectories, their possible future trajectories. Car lanes, traffic signal etc.

Surely one would be able to just look at that and decide, should I merge or not. It’s a way better view than side mirrors.

Basically I just want a 360 object view around the car on a heads up display as I’m driving. That would augment me as a human to be a better driver. Also alert me when it’s likely to be dangerous.

Basically the blog is saying, Waymo driver has superhuman eyes.

I understand what the blog says; I just have experience that learning to park via HUD is a very different game than looking directly; and that's a low-speed, low-object-count endeavor. Looking at possible intents might be overwhelming; but I trust that much can be achieved with training.

I'm still convinced that prime majority of ML, and "self driving" is plain fraud to get money out of investors, big businesses, and fat employers.

The best proof is that even with billions spent, the best contenders got marginal improvement over what Mercedes had in 1990 with discreet algorithms and no neural/fuzzy logic

Doesn't drive in anything, but the best conditions, and needs human intervention every 5km unless a passenger is ready to wait few minutes every time it has to pass a complex intersection the safe way.

I think them finally employing a mm-wave radar this time is an evidence to them backing off to more "dumb" and sure to work solution as a last line of defence once their fancy imaging/neural algo fails.

The question is: if after all they weren't able to make their neural net algorithm to best the "dumb" approach, why use that fancy "artificial vision" to start with?

According to their latest Californian DMV filing, Waymo's disengagement rate was 0.076 per 1,000 miles.

This means for every disengagement, they've driven for 13,158 miles on average. That's ~1000~ EDIT: 4, thank you; orders of magnitudes from your "5km".

Could a Mercedes in the 1990s drive itself for 13k miles without human intervention?

> This means for every disengagement, they've driven for 13,158 miles on average. That's 1000 orders of magnitudes from your "5km".

That's four orders of magnitude, not a thousand.

Thank you for the correction :)

FWIW, 1,000 orders of magnitude is an order of magnitude larger than a googol.

Also where these miles happen, and how long they take relative to a human driver (1.1x?) matter quite a bit.

edit: I've never shared a road with a self-driving car but I can't imagine I'd like it. I'm extremely patient with fellow meatbags (I work a lot of customer service, and enjoy it), but I'm not at all patient with software thinking it's smarter than it is. I also think people tend to underestimate how much "car-body language" is involved with driving as well, but I suppose that's the primary thing self-driving cars are learning.

Self-driving cars can also burgeon a surveillance state depending on how the fleet is implemented, particularly around what will be required for client-side redaction (pedestrians, license plates, windows etc). Keeping raw data locally for a period of time is fine in case of crash/malfunction, even backing it up encrypted with ephemeral keys can be well and good. But unfettered collection and access to unredacted data—perhaps something Google is drooling over—needs to be stopped cold.

To be fully pedantic, 1000 orders of magnitude (10^1000) is really 900 orders of magnitude larger than a googol.

10^1000 / 10^100 = 10^900

Woah, cool thank you.

I don't trust that this number is proving what you're saying it does. It's a meta point, but I found myself trusting such "data" less and less.

"Do you have data?" or "Do you have a peer-reviewed paper about it?" is an Internet meme at this point. Bringing some numbers or DOIs is table stakes now, in anything but dumbest of discussions.

These values - "0.076 per 1000 miles", "13158 miles on average" - have a history behind them. They're pulled out of a set of test drives. But how did the drives look? Was it a collection of similar drives, or a bunch of longer drives with significant disengagement plus a lot of short and easy drives to lower the average? To understand the true meaning of the numbers, you'd need to see (among other things) the distribution of distances covered by test drives.

I wish we had a way to require and embed context with the data, inline with the data. So that when I see "0.0076 per 1000 miles", I know it's:

  SELECT 1000
       * (SELECT sum(disengagements) FROM drives)
       / (SELECT sum(distance) FROM drives)

And so that I could click around and turn it into e.g. a probability distribution of:

  SELECT distance FROM drives

Or view a cumulative distribution of disengagements per distance:

   SELECT t1.distance, SUM(t2.disengagements)
   FROM drives t1
   INNER JOIN drives t2 ON t1.distance >= t2.distance
   GROUP BY t1.distance
   ORDER BY t1.distance

Now the problem usually is, this data is often not available - people release aggregate summaries instead of full data sets (whether that's the case with Waymo, I don't know - I didn't check, because I don't care about this particular case). But it should be made available, because point summaries of distributions are the easiest way to derive bullshit conclusions (and/or intentionally mislead people). And conversely, I don't trust point summaries unless I either a) review the underlying data (or at least know the shape of its distribution), or b) can trust that the person giving me the summary reviewed the underlying data (or knows the shape of its distribution) and isn't trying to mislead.

So there are two problems here: getting people to release more data, and creating tech infrastructure so that this data can be explored inline (preferably with nice UX that doesn't require user to type in SQL).

</end-of-meta-rant>

13000m ~> 20000km

    Order of 200000: 10^4
    Order of 5: 1^0

Conclusion: difference in order of magnitudes = 4 (not 1000)

There are endless ways to game that number. The biggest systematic bias of all must be simply California weather.

That's a pretty clever technology though right? Because ordinarily you'd expect Waymo cars driving in say Phoenix, Arizona to have er, Arizona weather?

More seriously, once you switch to "Actually I think in a snow storm this car isn't better after all" we're back to God of the Gaps style arguments which you can choose to lose incrementally if you want but it just looks kind of sad. It's definitely true that human drivers will set out in dangerous conditions and sometimes arrive safely. If, as Waymo promotes, the idea is self-driving cars are safer they just won't go in those conditions because it's unsafe.

This is a debated number, with people telling that they don't count anything, but actual "stall" when the system stands still and can't do anything, as disengagement.

And accounts of stalled Waymos in the middle of the road are too frequent to believe them

Disengagement is whenever the human test driver has to intervene. The car stalling in the middle of the road would only be counted as a disengagement if it doesn't recover on its own.

> And accounts of stalled Waymos in the middle of the road are too frequent to believe them

You gonna provide a source for that?

> The best proof is that even with billions spent, the best contenders got marginal improvement over what Mercedes had in 1990 with discreet algorithms and no neural/fuzzy logic

Can you elaborate on that? I haven't found much on that, e.g. w.r.t. the Mercedes system handling other vehicles, pedestrians, street signs, unexpected situations (roadworks, children playing, etc.) - from what I was able to gather, the two systems aren't even remotely comparable.

Edit: Mercedes was able to go on a highway, follow the lane, and keep distance to the car in front. Claiming "the best contenders got marginal improvement" is heavily biased at best - it feels like an attempt to justify an outdated opinion, rather than having facts inform it.

They haven't been relying solely on "fancy imaging/neural algo" - every self-driving car since the DARPA challenge has used LIDAR, and a good chunk of the self-driving R&D funding has gone into making those cheaper. The LIDAR they're bragging about in this blog post is them adding four more sensors to the car for short range, on top of the one they already had on top.

mm-wave radar gives them better sensing for long ranges and in bad weather, but isn't a fundamental change in approach from their already-sensor-heavy tech.

> them finally employing a mm-wave radar this time is an evidence to them backing off to more "dumb" and sure to work solution as a last line of defence once their fancy imaging/neural algo fails

AFAIK from little experience in the industry interpreting Radar images is not easy nor dumb

For example read here:

http://homepage.tudelft.nl/e15f9/pdf/Uysal_MAXWELL_19.pdf

> what Mercedes had in 1990 with discreet algorithms and no neural/fuzzy logic

Do you have any more information about it?

I could find only abstract info that the system existed: https://www.autoevolution.com/news/a-short-history-of-merced...

Full self driving is probably wishful thinking for the next decade at minimum. That said, even just getting to a point where you could use a human+tool combo to reliably shepard a second truck to a depot at the outskirts of a city would have profound logistics implications.

Reply on news.ycombinator.com