Catch up on stories from the past week (and beyond) at the Slashdot story archive

 


Forgot your password?
Close
typodupeerror
×
Transportation

Waymo's New Robotaxi Will Feature Fewer Sensors To Help Lower Costs (theverge.com) 88

Posted by BeauHD from the who-needs-em dept.
Waymo has unveiled its sixth-generation robotaxi, an electric minivan made by Chinese automaker Zeekr. While the company claims it's more advanced than previous generations, it features fewer sensors to help reduce costs. The Verge reports: [W]ithin its high-powered computer, it contains all the learnings of the previous five generations of Waymo's autonomous vehicles, meaning it won't have to do as much real-world testing as past models before it can be rolled out to the public. But looming over Waymo's assertion that its new robotaxi will be cheaper to produce is the possibility that it could also be subject to costly new tariffs against Chinese-made electric vehicles. Earlier this year, the Biden administration said it would quadruple tariffs on EVs from China to 100 percent, from the current 25 percent, as a way to "protect American workers and American companies from China's unfair trade practices." [...]

Waymo says the sixth-gen robotaxi will feature a streamlined sensor suite of "16 cameras, 5 lidar, 6 radar, and an array of external audio receivers (EARs)." These sensors will help provide "overlapping fields of view, all around the vehicle, up to 500 meters away, day and night, and in a range of weather conditions." That's the equivalent of over five football fields of visible range. Waymo's use of multiple sensors is important for redundancy, in which multiple sensors and cameras can ensure the vehicle can continue to detect and respond to its surroundings if something fails. It's unclear where and when the new sixth-gen robotaxis will first appear. "Waymo currently operates in Phoenix, San Francisco, and Los Angeles, with plans to launch commercial service in Austin, Texas," notes the report. "The company has been manually testing the Zeekr-made minivans on public roads, with the goal of adding them to its commercial fleet sometime soon."
This discussion has been archived. No new comments can be posted.

Waymo's New Robotaxi Will Feature Fewer Sensors To Help Lower Costs More

Waymo's New Robotaxi Will Feature Fewer Sensors To Help Lower Costs

Comments Filter:

  • This makes sense if you think about it. (Score:3)

    by mmell ( 832646 ) on Monday August 19, 2024 @08:49PM (#64719636)

    If their driving algorithm can achieve the same level of correct, safe operation with less sensors, using less sensors will make the system faster and more responsive overall.

    You still have to trust the software - making defective software faster just gets you to the accident site more quickly.

    • Re:This makes sense if you think about it. (Score:5, Informative)

      by JustNiz ( 692889 ) on Monday August 19, 2024 @08:53PM (#64719646)

      >> If their driving algorithm can achieve the same level of correct, safe operation

      As a Phoenix resident (one of the cities where Waymo operates), "Safe" and "correct" are not words I would use to describe Waymo's efforts so far. It needs to do much better than "same".

      • Yeah, but how many human drivers are much worse? The advantage of these driverless fleets is that every car gets updated with being able to handle new situations, while with human drivers they never properly learn and seem to forget. yes, we're still at the baby-steps part of driverless cars, but it is getting better and better each couple of months. The more of these cars on real roads the faster it improves. But there needs to be a central databasesystem with every single situation recorded and added when

        • Re: (Score:2)

          by JustNiz ( 692889 )

          >> Yeah, but how many human drivers are much worse?

          Then fix that problem directly, not try to work around it.

          • But that's the problem, there is no fix to the crappy human drivers other then banning humans from manually driving themselves (once driverless cars are a couple of generations along, like in 10 years or so, but that's also not possible as there are way too many older cars, so for the foreseeable future, like 30 years at least, we're stuck with all those older cars).

            • Re: (Score:2)

              by JustNiz ( 692889 )

              Yes there is. It's not a problem in the UK. for example, so why is it here in the US? (at least here in AZ).
              Because as far as I can see, the cops here in AZ don't even bother to stop people they see on their cellphones while driving (on my commute I regularly see them ignore drivers blatantly texting while driving etc). Actually the cops are sometimes the worst, including clearly using a laptop while driving. And even if the cops catch someone it's just a token slap on the wrist ($75-$149 fine). If it was t

        • This is a huge fallacy. For every mistake that is made by a human, it is not made by every human. On the other hand, every mistake made by an AV will be made with every AV. As more and more of these are on the road, every mistake they make is multiplied by a different amount. It is a really big difference.

          • For every mistake that is made by a human, it is not made by every human. On the other hand, every mistake made by an AV will be made with every AV.

            The corollary to your assertion is that every mistake not made by an AV, will not be made by -any- AV. Either they all make the wrong choice, or none do. Once we have identified a wrong decision, we can correct it for all occurrences. You can't say the same for humans.

            • No you have no clue how many other problems you create when you fix it.
          • Yeah, you're right, but the advantage is that every car gets a fix for this mistake, while with humans even the ones that made the mistake are bound to repeat it.
            • Yes one human is bound to repeat one mistake. How many mistakes does every Ev have inside them that we don't even know of yet. Care to analyze the mistakes that will come out of every EV when it hits blowing snow and ice? When you may create other mistakes by fixing one, you are not improving the situation at all.

              • Yes you might make another mistake by fixing one, but as I said, with decent automated testing there shouldn't be that much of a problem, especially if the automated testing system also has an AI part which might think of other situations itself.
                Humans make much more mistakes, especially in snow, rain, mist situations.
                No the driverless system won't be perfect, but it sure will be more perfect as a human driver. So even if the system is responsible for a death or two, it is still much safer as human drivers

                • If they let them on the streets thinking they could cause an accident, then they are premeditating that accident because they haven't tested enough. It is totally different if a human gets in an accident because a human doesn't have the luxury of running millions of possibilities before they go for a drive.

    • Re: (Score:2)

      by ls671 ( 1122017 )

      If their driving algorithm can achieve the same level of correct, safe operation with less sensors, using less sensors will make the system faster and more responsive overall.

      You still have to trust the software - making defective software faster just gets you to the accident site more quickly.

      Yeah, putting less of those useless components in is always a good idea and a guaranteed recipe for success and cost savings!

      Boeing tried something similar by putting less bolts on their airplane doors to reduce costs and it worked perfectly well.

    • I'm guessing that they can adjust the algorithm so that the outputs are pretty much the same for both the original and cheaper sensor setup. I hope that smart people are determining what "pretty much" means.

    • If their driving algorithm can achieve the same level of correct, safe operation with less sensors, using less sensors will make the system faster and more responsive overall.

      You still have to trust the software - making defective software faster just gets you to the accident site more quickly.

      Not all sensors are of equal importance. Generally the sensors detecting objects in the current path are more important, and objects closer to the car are more important than objects farther away. The trick is to prune away (or at least cheap out on) the sensors that are less important. Sort of like how intelligent pruning and quantization of AI models degrades functionality only slightly despite significantly lower resource usage.

    • It's interesting that they think they are ready to cut costs, instead of focusing on improving quality.

  • does this one kill people or just annoy them? (Score:3)

    by ebunga ( 95613 ) on Monday August 19, 2024 @08:59PM (#64719660)

    I can't keep my self-driving car companies straight. Is this the one that couldn't figure out how to keep it from killing people with a full sensor suite, or is it it the one that can't figure out how to park and keeps honking outside apartment buildings all night long?

  • Not ready for prime time. (Score:4)

    by msauve ( 701917 ) on Monday August 19, 2024 @09:02PM (#64719662)
    >Waymo currently operates in Phoenix, San Francisco, and Los Angeles, with plans to launch commercial service in Austin, Texas

    Let us know when they can drive through a Buffalo, NY snowstorm.

    • Let us know when they can drive through a Buffalo, NY snowstorm.

      Human drivers don't do so well with this either.

      • Re:Not ready for prime time. (Score:5, Funny)

        by msauve ( 701917 ) on Monday August 19, 2024 @09:42PM (#64719744)
        Did you ever notice - in a snowstorm, everyone who drives faster than you is an idiot, and everyone who drives slower than you is an asshole.

        • Precisely.

          I live in the south, but my wife is from Michigan. Michiganders like to tease southerners because they don't know how to drive in snow. What I learned though, is that northerners are just used to being in the ditch.

        • Re: (Score:2)

          by e3m4n ( 947977 )

          I thought that applied to all traffic. You should see Atlanta.

        • Re: (Score:1)

          by markjl ( 151828 )

          Did you ever notice - in a snowstorm, everyone who drives faster than you is an idiot, and everyone who drives slower than you is an asshole.

          Credit for the quote = https://en.wikiquote.org/wiki/... [wikiquote.org] -- last bullet point.

      • Human drivers don't do so well with this either.

        Ok by all means lets compare a human driver to an AV in that situation!

        • Agreed, we should compare apples to apples. My only point is, if the threshold is "I'll wait until AVs can drive in a Buffalo snowstorm" that's a threshold that is unreasonably high, since humans don't do well in that situation either. If an AV can do *as well* as a human, that's a sufficiently high threshold for me, and AVs have largely reached that threshold already.

          • I grew up where it was cold. I have seen millions of people successfully drive in those conditions. I think I have only seen four accidents in my life, yet how many other cars must I have seen NOT getting in an accident?

            • You clearly don't live in Buffalo, which had nearly 10,000 traffic accidents last year. https://www.makeroadssafe.org/... [makeroadssafe.org].

              Or for that matter, any other city of any size.

              And most cars that slide into the ditch because of a snowstorm, are never reported as accidents. The owner simply calls for a tow, and that's that.

              • And how many automated vehicles in Buffalo to compare to?

                • My comment was regarding the fact that you've only seen 4 accidents. This is not representative of what is typical for people who live in the US, since the average person in the US is involved in 3-4 accidents in their lifetime. https://www.ddlawtampa.com/res... [ddlawtampa.com].

                  The math here isn't that hard. Even though AVs are a small percentage of all vehicles, we can use ratios to compare accident rates for AVs with human-driven vehicles. And while the data is evolving, current data shows that self-driving cars are now

                  • So the average person is involved in 4 accidents in their lifetime. They drive 14,263 miles per year on average. The average age to get a license is pretty close to 20 and I would say 65 is a conservative age to say that they stop. So they drive 14,263 * 45 = 641,835 miles and get in 4 accidents or 160458 miles without an accident. Keep in mind that is in rain, snow, fog, ice, high winds, on road, off road, navigating snow clearing equipment, navigating construction zones, etc etc. Automated vehicles

                    • So using your numbers, humans have about 6 accidents per million miles, AVs have 9. Even if the number for AVs is artificially increased by choosing roadways, it is also artificially decreased by Tesla's less-than-fully-ready self-driving mode, which accounts for 2/3 of all self-driving accidents. So as I read it, the odds are pretty even, between self-driving and human-driven cars. Close enough that I'll gladly risk my personal safety in a self-driving car, should the option become available to me.

                    • Ok you do you. I don't really want to be killed by a robot that could have been programmed properly in the first place.

                    • And I don't want to be killed by a human who might have been taught improperly, or didn't listen, or was distracted, or doesn't care, in the first place. And unlike machine bugs, human deficiencies aren't getting better over time.

                      I'm glad to see you've come around to (tacitly) admitting that the risk is about the same, otherwise you would no doubt dispute the math.

                    • I didn't address your comment about the risk being even because you didn't even have any numbers. "Tesla drags down the numbers". By an accident every 60,000 km? I doubt it. Please cite your references.
                    • Also, I have news for you... Just drive with a person who is a safe driver and you can almost totally eliminate your chance of being in an accident.

                    • Yeah, except that you have to share the road with other people who may be unsafe drivers, and you have no control whatsoever about that.

    • Re: (Score:1)

      by Hadlock ( 143607 )

      Isn't Buffalo the failed city that's been shrinking almost as fast as Detroit? I doubt anyone is targeting that market.

    • >Waymo currently operates in Phoenix, San Francisco, and Los Angeles, with plans to launch commercial service in Austin, Texas

      Let us know when they can drive through a Buffalo, NY snowstorm.

      It's fine if the cars can't operate in Buffalo snowstorms. Level 4 autonomy is not Level 5 but would still be extremely useful and life altering.

    • >Waymo currently operates in Phoenix, San Francisco, and Los Angeles, with plans to launch commercial service in Austin, Texas Let us know when they can drive through a Buffalo, NY snowstorm.

      Waymo is actually doing weather testing in Buffalo. I do expect they will learn a few things. As a past NYS upstate resident, I understand snow can be an interesting challenge (and I recall pulling out my tow chains more than a few times when driving my 4x4 truck around town and pulling the idiots out of the ditch in my community).

  • This is how optimization works (Score:3)

    by Tony Isaac ( 1301187 ) on Monday August 19, 2024 @09:28PM (#64719706) Homepage

    First, you get it to work, then you optimize. In the first iteration, they figured out they don't really need quite so many sensors. Good for them.

    • It still seems like a huge sensor suite. I know it is doing much more than Tesla can, but 80% of Tesla's issues could be solved with forward facing HD radar. The directional microphones are likely a good idea as well for emergency vehicle detection, and I can see value in a few more cameras.

      I struggle with what the LIDAR adds to the party; I could see forward LIDAR maybe offering help in some scenarios, but the HD radar seems to have too much overlap in these areas.

      • Re: (Score:2)

        by BranMan ( 29917 )

        Fog, I imagine, is what the LIDAR is for. Like our eyes, cameras can't see through fog, but LIDAR should cut through it pretty well.

        RADAR doesn't help too much - there are way too many things on the roads (like the occasional pedestrian or deer) that don't reflect RADAR well at all. For seeing cars and guardrails (assuming they aren't wooden) it's good. Carbon fiber bike plus rider? Eh, not so much.

    • First, you get it to work, then you optimize. In the first iteration, they figured out they don't really need quite so many sensors

      Eventually, just like Tesla, they will realize the only sensors required are cameras, just like for humans the only sensors required are eyes.

      • I thought the point of self-driving cars was to be better than humans. Our senses are pretty limited. I wouldn't mind having a RADAR/sonar/LIDAR sense or being able to see in other wavelengths.

        • I thought the point of self-driving cars was to be better than humans. Our senses are pretty limited. I wouldn't mind having a RADAR/sonar/LIDAR sense or being able to see in other wavelengths.

          The point is to seem better than humans at minimal cost. I mean, let's be honest, these initiatives are profit based first. Saving humans is a secondary concern, if even that.

      • And Tesla has the highest accident rate of any auto brand. https://www.forbes.com/sites/s... [forbes.com]

    • I don't think they have proven to anyone's satisfaction that it works.

      • I'm certainly convinced. Have you seen human drivers lately?

        https://www.newscientist.com/a... [newscientist.com]

        • Reduce the incidents you see by human drivers by the number of humans driving cars versus automated. Of course you will see more human mistakes.

          • The study already takes into account the difference in the proportion of human drivers, to automated vehicles.

            • Driverless cars seem to have fewer accidents than human drivers under routine conditions,

              I'm not even sure how they can know that, since severe snow/fog/rain is part of 'routine conditions' and to my knowledge automated vehicles don't drive in adverse weather conditions.

              One research roadblock is the “autonomous vehicle accident database is still small and limited”, says Abdel-Aty. He and Ding described the need for “enhanced autonomous vehicle accident reporting” – a major caveat echoed by independent experts.

              There is the real reason this study is bogus of course. They don't have a full database to go on. Results of automated vehicles are heavily weighted due to the fact that their routes are carefully selected.

  • we don't need to sweat the Chinese balloons flying overhead anymore. Their car sensors on waymo will gather all the data necessary. So weird they keep taking the routes near the military bases.

  • Living in Phoenix AZ, I've had the opportunity more than once to ride the Waymo.
    It's fun the first 2 times - the algo is pretty clever in some situations.
    It attempts to avoid left turns across busy arterials at uncontrolled intersections, which is a noble cause.
    Then I rode it to/from the airport, which is dead straight down 24th street from my neighborhood to the west end of the airport. It's not allowed (I think) on the freeway, which would be most direct. However, it ended up taking a somewhat circuitous

    • . On the way home, it added about 3 miles (to a 6 mile route). Dunno why

      For the same reason cab drivers do it, it's called "long hauling" and is used to boost the cost of the ride.

      • I don't think so because, I believe, companies like Waymo give you the price up front and have an incentive to take the most efficient route. Lyft and Uber certainly do this. It's one of the many things that makes the Uber/Lyft customer experience better than the taxi experience. I don't like the ride hailing companies as they are basically illegal taxi services that should have been shut down. But I also have only bad memories of taxis so I'm sympathetic to the people who welcomed Uber/Lyft with open a
    • Reports are that Waymo only drives on streets where it feels confident. (And yes, that is foolishly anthropomorphizing a computer, sorry Dijkstra). It's a smart move, however, it can add quite a lot of time to the ride.

  • Now more Crashy!

  • It's just progress which why there are less sensors needed. Also the new sensors used are improved over the previous sensors. With all the experience from previous models and with the advancements in processing power, it makes sense that fewer sensors can be used.

  • Imagine a humanoid robot that can carry bags, connect an EV to a charging station, clean the vehicle inside and out, satisfy your wife, etc.

    It could sit in any driver's seat and adapt to the vehicle, using only the sensors mounted on its own body.

  • Capitalist Tech Company Will Kill More People To Help Lower Costs
  • We need laws to eliminate driver-less cars. You neither plan for every roadside contingency, nor pedestrians, kids, pets, bicyclists, and motorcyclists.
  • Does anyone know whatever happened to super cheap solid-state LIDAR?
    It has been announced quite a few times, and should make LIDAR prices plummet, in theory. But all iages of robotaxis I see still seem to have classical LIDAR sensors which seem to be very expensive?
    Also, Musk is an ass-hat, but I like Tesla's vision-only (or vision-first) approach. If a human can do it, so should a computer using a couple of cameras, after all. And cameras are super cheap. But I hope they at least plan to be able to augme
  • People will die so that even more wealth can be taken away from blue-collar workers.

  • So max sensor range is 500m? Sounds a bit crappy. At highway speeds you go from the 'dead kaiju blocking the highway being totally undetected' 15 seconds from impact to 'emergency braking must begin now' at 2 seconds from impact.
    The only chance this has of working is if the electronics, programming staff, programming languages, programming methodology, process management and project management is utterly separate and distinct from that of any other computing device that anyone has ever used in the entire hi

    • Re: (Score:2)

      by BranMan ( 29917 )

      500m is plenty, and you know it. As you pointed out, that gives you more than 15 seconds of lead time. At 80 mph, which is well over the speed limits most everywhere.
      Recommended lead time while following another vehicle is 3 seconds. And that's for a human, with human reaction times - electronics should do much better.

      So, 15 seconds out, sees nothing, At 12 seconds away, sees something in the road. Not sure what it is. Let's let off the gas and coast. Speed is dropping right now. At 10 secon

Slashdot Top Deals

The reward for working hard is more hard work.

Close