Hacker News new | past | comments | ask | show | jobs | submit login
Waymo: Bringing 3D perimeter Lidar to partners (medium.com)
136 points by Fricken 14 days ago | hide | past | web | favorite | 61 comments

I work as a lead autonomous vehicle engineer and have designed a sensor suite.

Here's a few thoughts:

* They say it has a 95 degrees vertical FOV and 360 degrees FOV. The vertical FOV is awesome actually. A Velodyne VLP-16 has roughly 30 degrees FOV. We need the vertical and horizontal angular resolution to make any first-take judgements though.

* One overlooked feature I want to point out is that the Minimum Range is 0. Many lidars perform well at far distances but struggle to see objects very close (<1 meter away). This poses a safety issue as someone or something is effectively invisible to the robot if they're too close. Of course you can use cameras to handle this case but this is a big win for the Waymo hardware team!

* They're going to be 15k+ (gut estimate)


By the way, if you want to to become an autonomous vehicle engineer I put together a resource guide here: https://becomeautonomous.com

Thanks for the excellent resource Kyle!

The self-driving industry has not progressed fast enough to get LIDAR prices down. Continental announced a good automotive flash LIDAR several years ago.[1] They're a big auto parts company; they need an order on the scale of typical auto parts to start volume production. The solid-state solutions are potentially cheaper, but involve custom ICs, which need volume to get the price down. (Continental's technology is from Advanced Scientific Concepts, which makes high-end LIDAR units. The Dragon spacecraft uses one for docking. Costs about $100K. So it's not vaporware, like some of their startup competitors.)

[1] https://www.continental-automotive.com/en-gl/Landing-Pages/C...

Well there's Livox with a $600 unit: https://www.livoxtech.com/ Initial reviews[1] on the internet show it looks pretty decent for the price. There are certainly better sensors, but I would argue we have indeed entered the sub-$1,000-lidar-phase of the SDC era. As if Andrej Karpathy needed anything more to worry about.

[1] https://www.reddit.com/r/UAVmapping/comments/aimnik/livox_mi...

There's no information about price. I heard that lidar's price was a massive problem for their mass adoption and the reason why Tesla try not to use them for self-driving cars. If its price would be signinficantly lower than other lidars, it would be a real breakthrough.

Any idea about its price ?

Waymo isn't disclosing price. Part of their reason for taking this initiative is to find a market that justifies scaled production of their devices, as a means of driving the costs down. Everyone in the Lidar bubble hopes to someday have Lidar that is both cheap and good, and such a product is likely years away, for Waymo or anyone else.

This announcement reads like a 'Plan B' for Waymo's lidar team. Waymo's talk of buying 20k Jaguar Ipaces and 60k Chrysler Pacificas for robotaxi deployment is indefinitely on hold, and if that was playing out the way they wanted it to, then Waymo's lidar production team would have more important things to do than fishing around for a secondary Lidar market to sell their wares to.

Can you cite a source for vehicle purchasing on hold?

I'm not saying they cancelled their puchasing order. When they announced those purchasing orders a year ago Waymo's PR was singing a different tune, they've missed a lot of targets between then and now. It's hard to paint a plausible scenario where they deployed even 10k of those robotaxis over the next few years, let alone all 80,000 of them.

Was there any timeline given when that order was announced? It could have been over five years, for all we know.

No they never gave a timeline. Waymo announced they had bought or were going to buy 600 Chrysler Pacificas, in summer 2017, then they later announced plans to buy 2,000 more, and then later announced plans to buy 60,000, but currently they still don't have all of those original 600 in operation. 5 years to get all those vehicles earning money is sounding optimistic to me. The complexity and the number of complications and hangups Waymo has to deal with just keeps piling on.

An interesting comparison (maybe) is the LIDAR unit that was developed for the Neato XV series robot vacuum cleaners.

When it was first announced, and the research paper was published showing how it was built and functioned (essentially a 2D spinning parallax distance measuring system using a laser and linear photosensor array), it was thought that the system would be very cheap to make, and would eventually sell to hobbyists for around $25.00.

Unfortunately - that never happened.

I don't know what it ultimately cost Neato to make; it was likely inexpensive in quantity, but it was never sold to hobbyists or robotics researchers as a standalone product. Instead, if you wanted one, you had to either purchase and cannibalize one of the robots, or otherwise obtain one as a replacement or "pulled" component. Even then you would spend much more than $25.00 to get one.

Eventually they started to show up on Ebay; at one point I was able to purchase the units for around $35.00 at the cheapest; these were "used pulls", some needed new motors, pulleys, and/or drive belts (all parts easily found as well) - so that pushed the price up a bit.

Lately, the prices on Ebay have settled to around $75-90.00 for refurbished pulls, or replacement new units.

Interestingly, similar units have also appeared on "Chinese clone" vacuum robots, and you can purchase the LIDAR units they use separately on sites like Alibaba, etc. But even there they don't sell for anywhere near $25.00 each.

I am not sure where the discrepancy for this device lies; maybe the quantity just isn't there, and honestly $100.00 for a 2D LIDAR (granted, indoor only and very limited range) is fairly inexpensive, but even so I don't see many people using them on their hobbyist devices. Most research robotics continue to use the more expensive SICK and Hokuyo 2D LIDAR (which are time-of-flight based).

I think it is more likely that if LIDAR is used, it will be part of a large suite of sensors, and likely 2D only, unless the price of flash depth LIDAR devices plummet.

I'm not an expert, but I personally think that computer vision based approaches can work well, especially combined with other sensor systems (LIDAR, RADAR, ultrasonics, etc). NVidia has shown that vision-based approaches can be made to work, and others have done similar work that show the approach to be viable. I'm a bit biased on this because I used a simplified version of NVidia's approach to pilot a simulated vehicle in an online MOOC I was a part of a few years back (I had to simplify the CNN to make the model fit the memory I had in my GPU at the time; I had tried to use a batch training method but wasn't able to get it to work properly before my deadline - even so, the simplified model trained and generalized quite well with the data I gathered).

I left a comment on the article asking for a "safe overestimate". Should at least be able to say it will be under $N.

> Now, we are making these sensors available to companies outside of self-driving

They are explicitly limiting this to use cases that don't compete with self-driving, so it is likely bound by licensing and non-compete agreements.

So don't expect seeing one of these sensors in a competing self-driving vehicle any time soon.

I think if you wanted to get into the self-driving car arena and were a startup on a budget, and you wanted to play with 3D LIDAR - then the place to hit up for one would be Ebay.

Lately I've seen used Velodyne 3D LIDAR units being sold on the site for (compared to new) dirt cheap. Granted, much more expensive than used SICK or Hokuyo 2D sensors - but that would be expected.

The fact that they are being sold on Ebay at all is a fairly new and interesting development; the question becomes "where are they coming from"? That is, what is the secondhand source they are being pulled from.

In the case of SICK and Hokuyo sensors, it has generally been industry or research. For Velodyne sensors, they were explicitly made for self-driving vehicles (though they've found other markets since).

It makes me wonder if one or more the established self-driving vehicle companies are shedding the tech in favor of some other system...?

The Techcrunch article ballparked the price below $5000. They had already reduced price by 90% about an year ago and insiders say that further reduction has been achieved since then. ( written in the TC article)


Thanks, I referenced this quote and TechCrunch article in my comment on the medium article.

You're welcome. May I have your article link ?

I would imagine Waymo wants to sell them to other companies to reach better economies of scale and bring their own prices down.

It'd be interesting to hear from someone who knows specifics of LIDAR manufacturing, though.

Brainstorm thread: What are some novel ideas outside of self-driving that the Waymo Lidar Sensor could be used for? And what could be done if it goes below $1,000 (e.g. video games)?

Automated solar calculator. Point to a roof, get a 3D model and use some model to estimate how much of roof you can cover with panels.

Guesstimate the build cost + total solar input per year. Get instant payback number.

Use LIDAR to establish some sort of canonical clothes size format? Not just S, M, L, XL that only works for your region. Maybe you don't even need LIDAR, but I'd love personalised clothing.

If it's cheaper than thermal cameras - perhaps some use for lifeguards, especially in night?

Google has already built a great solar calculator using satellite imagery, at https://www.google.com/get/sunroof

Technically, it's not satellite, but aerial imagery from multiple angles. Geometry is reconstructed with pose estimation and SfM (the open source Ceres solver originated from this and other work).

Your first one could probably be done with photogrammetry (plus software), and the second could be done with a 3D scanner. For the third, the beam spread probably wouldn't pick out a person at that distance away from shore.

All of these use cases could be served - perhaps even better so - with standard cameras and 3D algorithms (e.g. ARKit and the likes).

You could prototype the solar calculator yourself with any modern mobile device in a trivial amount of time.

They say they can record multiple returns per single pulse. This can really useful for forest mapping (e.g. on a drone).

For video games maybe but I'd always try to avoid looking at those, even if LiDARs are eye safe it's not fun staring at them.

Recording multiple returns has been common in aerial mappers for years. It's called "first and last" in the industry.[1]

[1] https://gis.stackexchange.com/questions/142443/what-are-lida...

Wow, detailed forest mapping! What are some things that would be doable with detailed forrest maps?

Tree structural models: https://www.mdpi.com/2072-4292/5/2/491 Forest inventory: https://www.sciencedirect.com/science/article/pii/S092427161... And much more :) Source: I am doing my masters in lidar for forestry and robotics. Feel free to ask me any questions.

Perhaps mapping tree growth in response to some environmental changes, or detecting illegal logging, or measuring the impact of wildfires, or, maybe if it does a good job of detecting animals, mapping the migrations/populations of endangered, larger mammals?

Alternately, a cheap way to produce amazingly accurate video game maps of your neighborhood.

You can also mix in signal from multispectral cameras to get a sense of the health of the flora.

Great stuff here, thanks for those ideas and insight!

CO2 credits

Traffic signals. We shouldn't have to wait at a red light if there aren't cars coming the other way.

Much cheaper cameras may be fine for that though...

You already have 2 generations of MS Kinect. It can already stream RGB-D images, lidar provides no color information. There is no killer application for cheap lidar, there are many cheap devices covering <10 meter distance range with great resolution and their adoption is still extremely low.

It's great that Waymo has decided to start selling bits of their tech. That's in contrast to other Alphabet-developed hardware like TPU (for training, not inference). I wonder whether 'for selected partners only' is going to be temporary or permanent restriction.

> Now, we are making these sensors available to companies outside of self-driving — beginning with robotics, security, agricultural technology, and more — so they can achieve their own technological breakthroughs.

It will likely come with licensing and non-compete agreements.

It would be as bad as lack of right to repair. If it happens , then I hope someone like Intel does the same to them.

I'm not sure TPU is a good example. That one is already super Google specific and difficult to externalize except through a hosted service. Google's internal infra is pretty complex.

It's not super Google specific if it is supported by tensorflow. Yes there may be some obstacles to using it to in your home pc, but there should be at least a paid way to use it if you want in your own server.

> It's not super Google specific if it is supported by tensorflow.

That's simply not true. If framework X supports <client that connects to complex hosted service Y> then "using it in my home PC" isn't automatically possible.

Can someone give me some other common examples of will-not-compete clauses in sales agreements? (This post implies they have one.) I'm drawing a blank.

A lot of web APIs, especially for B2B SaaS, come with these. Uber's does, too.

Anyone know if this is a solid-state unit?

I've heard Velodyne has a new solid state array in the works, but no idea how it compares to something like this. At CES this year there were quite a few vendors with various flash LIDAR tech.

It's not solid state because it's moving. Solid state usually uses an IR flash technology. :)

Any speculation about specifications of this device? How many points per second does it produce? No info in the article and on Waymo’s page. What it can do more than current lidar from Sick or Teradyne?

Way more. They say it has a 95 degrees vertical FOV and 360 degrees FOV. The vertical FOV is awesome actually.

A Velodyne VLP-16 has roughly 30 degrees FOV. We need the vertical and horizontal angular resolution to make any real judgements though.

Interesting that they designed it with 360 degree horizontal field of view when they are using it on the bumper of the car.

Seems like a good time to mention Waymo has a valuation of 75 billion dollars and has yet to make any revenue. Maybe they will start to bring in some money!

They do have revenue (although the cars have safety drivers).

I imagine there's some sort of "bait and switch" here where Google gets what it has always wanted and most car makers have refused to give it: access full car data and telemetry.

This sounds much more like illegal antitrust behavior as reported by independent media: https://www.theverge.com/2019/3/6/18252561/waymo-sell-lidar-...

"Select partners" can mean "those who won't compete with us".

Don't really understand how this is anti-trust behavior. Is a company not allowed to choose whom it's selling its products to?

Generally, no. Usually the right of first sale ensures anyone can buy a product, if not from the original manufacturer, than from a reseller. Extremely tight distribution chains designed to prevent a competitor from getting a hold of your product is pretty darn rare.

And when it's a monopoly that's already repeatedly demonstrated monopolist behavior, it's even less permissible. So they have a patent monopoly on the hardware, already shown they'll fight in court to protect it (re: Uber), and will prevent anyone from buying that hardware if it's a potential competitor to their service-based business. Their choice to sell the hardware makes it a separate market than their ride-sharing service, and one could argue a violation of antitrust law if they're blocking competitors from one market from buying their product in another.

There's no such thing as a "patent monopoly", and there are multiple competitors in this space (4? mentioned in the article). The idea that there's a meaningful monopoly here is strange, to say the least.

Unless you're actually trying to argue that "warehouse lidar produced by Google" is a meaningfully different market than "warehouse lidar", which completely contravenes both written law and precedent.

Also the right of first sale is trademark/copyright law, not contract law. Companies are absolutely allowed to create non-trademark agreements, and they do all the time. Think software licenses.

> Also the right of first sale is trademark/copyright law, not contract law.

And apparently patent law, you can't use your patents to force someone who legally acquired a device to not do something with it.

So, if one legally got their hands on one of these things there's not a thing waymo could do to them if they were "misusing" it.

The blog post itself says they will sell only to "select partners." The Verge's article doesn't add anything, and it doesn't claim it's illegal.

It's a safety-critical component. All they have to do to prevent competitors from using it would be to only certify its use in limited contexts.

Calling The Verge “independent” is a bit missleading.

True, but I'm giving the benefit of the doubt, since they're being critical of Google here, despite their incredibly cozy relationship with Google. Numerous Google executives have done exclusive interviews particularly with The Verge, all the way up to Sundar Pichai, so there's definitely some question whenever they praise Google if there's a conflict of interest or incentive for them to give Google a positive spin.

I assume, of course, you were likely trying to make some sort of "iVerge" claim here, but the reality is, The Verge gives Google incredibly favorable press coverage as well. At the end of the day, The Verge is a business that's in it for clicks, and exclusive interviews and insider access equals clicks. So The Verge will shill for just about any company that can give them scoops, including both Apple and Google. Both companies are notoriously brutal to sites that write critically about them, and The Verge avoids heavily criticizing either in most cases, as to not jeopardize their access.

Why is that?

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact