Seems like the Mid and Tele are using Risley prism beam steering, which would explain the hypotrochoid scanning pattern. [0]
They try to spin the scanning pattern as being an advantage as you get a more dense sampling if you point at it in a certain direction, but I'm not convinced. It is only an advantage if your device sits on a tripod. On a moving car, more predictable, structured patterns such as the Ouster OS-1 [1] make it much easier not only for deep learning, but also for SLAM (e.g. the LOAM algorithm [2] extracts feature points row by row, i.e. by ring). For a moving car or drone, any scanning pattern will sweep into a dense 3D point cloud anyway.
The pricing seems to be competitive if you only need a small field of view.
For 360 degrees, it takes four Mid-100s, which would cost $6000, or twleve Mid-40s, which would cost $7200, to obtain the same field of view coverage and the same number of points (1.2 M points per second) as a single Ouster OS-1, which is available to university researchers for $8000. However, buying a single Ouster OS-1 saves you the headache of extrinsic calibration between many lidar units, and the Ouster OS-1 only draws 14 W whereas four Mid-100s would draw a whopping 480 W. Four Mid-100s also weigh twenty times as much as one Ouster OS-1. For high density drone mapping, the Ouster OS-1 seems like a much better choice.
That said, the Livox does have a range advantage over the Ouster OS-1.
You think that Zhang odometry paper is good, you should see the stuff his grad student came up with for his thesis[0]. The results look almost as good as the figures from the Ouster blog post! ;-)
Where are you seeing the 480W power requirements? The manual says average power for Mid-100 is 30W and peaks at 60W. I presume the peak power is at low temperatures when the motor is difficult to start, if it is Risley prism.
Don't trust this thing. AFAICT, there's not a single clear photo of the actual physical product. Everything is a rendering/CGI/dark lighting and I can't tell if it's real.
If you had a working, quality thing, you'd expect they'd show a clear image. Warning!
It's pretty hilarious; if you click on the "buy" button you have to agree to this:
https://www.livoxtech.com/3296f540ecf5458a8829e01c
(aka don't steal our stuff and don't) " (j) disclose to the public the results of any internal performance testing or benchmarking studies of or about the Productswithout first sending the results and related study(ies) to LIVOX, and obtainingLIVOX’s written approval;"
Yup. Once one does it, especially if that one is the market leader, then their competitors tend to do it too, because otherwise the one that did it would be able to go ahead and publish benchmarks showing it beating the others, and they would not be able to counter.
Someone, I forget who exactly, tried to limit their restriction. It said something like you could not publish benchmarks of their product against your products that had benchmark restrictions unless (1) you published full specifications and configuration information so others could reproduce, and (2) you gave everyone permission to do so.
Usual caveat that you'll only get headline performance at 80% reflectivity, maybe a number plate? Also the IP67 rating system doesn't cover the cooling system (only IP55).
Hokuyo 2D systems start from about $1k and are well regarded in researcher. SICK systems were historically used on research autonomous vehicles (eg the grand challenges), but they're far too heavy for drone use.
Edit - I thought it was 1D initially, apparently the scan pattern is circular, think spirograph (or a lissajous). You'd normally do this with galvos, but this seems to be solid state.
They need a market first, no? They can't just produce 100k of them and hope to sell. To me it means that thy have all the know-how plus parts and manufacturer ready to go at this price part.
Of course the could be lying...er...overly optimistic and hope to get them ready after signing a contract, ala Bill Gates with DOS.
Depends on what market they're trying to get into. For consumers they might be away to get it. But for industry, there is an expectation of inventory which is why it's tough sometimes to have a hardware startup. Even then, it usually makes sense to make them in batches of 1000+ rather then made to order due to economy of scale.
Retroreflective road signs get close to 100% light returned and are a problem for most LIDARs because they blind the sensor. Some companies are working on car paint with higher visibility for LIDAR systems (https://automotive.basf.com/news/read/improve-lidar-detectio...) but that seems the wrong approach to me. The LIDAR needs to work with legacy infrastructure and nature.
I'm currently thinking through and costing out converting a classic pickup to EV / more modern control systems. It may not last 30 years but it's not getting self driving anytime maybe ever... and all the stuff we're making now is legacy.
Just pricing out some 98-'02 hondas and acuras for daily mileage gobblers as well, they're insanely cheap and pretty reliable. This is a 20 year old car that is in high availability very very cheap ($1700 in seattle), and I hvae no complaints about driving.
30 years may be optimistic except that the kids aren't learning to drive as much being the down pressure.
Super cool that this stuff is getting into pricing ranges reachable by mere mortals.
That's helpful. The scan pattern looks like they have an oscillating radial scanner plus a rotational scanner. I tried to build something like that once, using an military surplus Bulova resonant scanning mirror for the radial axis and a prism from a laser printer for the rotating axis. 1990s, before you could buy anything useful in a LIDAR. Never got very far.
You can do this, but unless it's super cheap, why? Price/performance isn't that good.
in the video, they mentioned there is no moving electronics parts. NOT no moving parts. I think they mean no slip-ring kind of easy-to-fail parts, which is necessary to connect the power and signal for rotating electronics.
Exactly, if it would of had solid state tech, I’m sure they would of mentioned it, that’s why I was unsure. It’s interesting to see the raw, competitive, old snake oil tactics in such a bleeding edge, socially useful, new technology. I cry a little for humankind.
Slightly off topic, but still some questions that have been bothering me for a while:
LiDAR for autonomous vehicles is cool and all, but how will this perform if there is not the occasional car making test runs through town, but when all vehicles on a busy road or intersection are equipped with a one or more units?
I understand that both radar and ultrasonic suffer from interference, so will this be different?
Are there plans for doing any kind of (cooperative?) frequency or time division to avoid interference between cars?
How DOS-able would LiDAR be? Would I be able to bring traffic downtown LA to a halt in 2040 by bringing my portable LiDAR jammer?
People don't worry about it because it just isn't a concern. The duty cycle of any particular lidar in any particular direction is very low and so for any two lidars they will have their beams pointed at the same place very infrequently. It happens but in order to, say, drive in the rain self driving cars need to be able to reject a certain amount of sensor noise. I've had a gagggle of robots using planar lidar, where the risk of interference is much higher, all work together in the same room together and not noticed any degradation in performance.
The reasons lidars are so useful it that they're very directional in a way that radar isn't. That means that an effective lidar jammer can't just emit enough to interfere when the lidar is pointed at it, it needs to cause every other object present to also emit that strongly. This is possible but will cause nearby people to catch fire, at which point it probably doesn't make sense to call the object a "lidar jammer".
> but how will this perform if there is not the occasional car making test runs through town, but when all vehicles on a busy road or intersection are equipped
This is called "jamming" and was a real problem for early radar solutions.
Curiously you well rarely see anyone mention this in the LIDAR space. For many of startups adressing this issue seems to be an afterthough.
One proper way to solve this is to use orthogonal codes in the scanning scheme. There are different ways to implement this. The complexity depends a lot on the used LIDAR architecture. It would not be surprising to see some of the startup being weeded out by that during real-world tests.
I've argued that your outgoing beam timing should have a few microseconds of random jitter, so that no jammer can synchronize to it. Military radars do things like that.
Yes, lidars have interference, but much less than radar and sonar. From my other comment [0]:
> Compared to other sensors, lidars are not that prone to interference because:
> * it only takes 1 microsecond to make a ranging measurement up to 150 m, so your detector is on for a short time
> * lasers only illuminate a small spot, and the detector is also looking at a similarly small spot, so it is unlikely for two lidars to point in the same spot
> Now, even if it does interfere, you may see a stream of random points pointed towards the interference source. This may happen if, say, you point a lidar directly at the sun, or if you have multiple lidars mounted on the same vehicle. Such random points are easily rejected as outliers and do not affect the vast majority of the scene. Most self driving cars (I hope) should have outlier rejection schemes that deal with outliers caused by this and other sources, such as snow, smoke, and so on.
Now, as for bringing traffic to a halt by bringing your portable lidar jammer... https://www.xkcd.com/1958/
> How DOS-able would LiDAR be? Would I be able to bring traffic downtown LA to a halt in 2040 by bringing my portable LiDAR jammer?
Even that will not be necessary, a splat of dirt on the sensor is enough.
I don't understand their obsession with LADARs. MM wave radars were successfully handling the same task in the industry for a few decades while being much cheaper and reliable.
NIH too strong. I feel calling that here is 100% appropriate and objective.
Radar is very useful but doesn't replace a lidar. Radars give very accurate readings of an objects distance and relative velocity but very poor angular resolution. For that reason they aren't much use for avoiding stationary objects since those are hard to distinguish from the ground, signposts, etc. At least until they get so close that the radar field of view lets you rule out those things.
They are really great at figuring out whether cars are braking, though, since cars are moving with respect to the ground you can just filter the ground out and with radar you can sense the change in speed directly rather than having to figure out the speed by comparing distance over time.
EDIT: Actually, I think automotive radars don't give any directional information at all since they don't use big dishes. I'm used to working with aerospace radars but even with those I wouldn't rely on one for car navigation.
> Radar is very useful but doesn't replace a lidar.
It does, and quite well. Without any additional processing, you easily get few cm accuracy, and sub-wavelength resolution is possible if you do. That's more than enough.
Latest automotive radars use electronic beam steering – not much different from what is used in latest missiles.
I'd say, companies opting for ladars for driving assists don't have good engineering expertise.
The brains processing the output certainly don't do that (emergency break logic is much simpler than automatic obstacle avoidance,) but nobody prevents making them do so by throwing more processing power on that.
Does such an agreement have a legal force (in US/EU)? I get the patent laws and not violating the IP, but I find it strange that just the curiousity of seeing how it works can get lawyers attacking you.
Earlier this year, the Tenth Circuit court upheld a preliminary injunction granted in favor of an electronics equipment manufacturer against a reseller of its goods in a trademark infringement action. In Beltronics v. Midwest Inventory Distribution, the reseller (Midwest) argued that it was able to resell the manufacturer’s goods based on the first sale doctrine. The court, however, disagreed with this assessment and ruled that the resellers violated the manufacturer’s trademark rights because Midwest’s sales caused consumer confusion.
Yes that's right. US courts have ruled that if you buy something, you have the right to resell it regardless of silly agreements like this one. You do have to be careful how you represent trademarks.
Isn't our perception of "reflectivity" essentially logarithmic? What we perceive as the "middle" between black and white only reflects 18% of light. Meaning anything that's darker than gray will be exponentially less reflective than that.
If you're talking about how the human eye perceives light or dark objects, then yes it can be more complicated. Especially when colour is involved. That said, what they're talking about here when they say 'reflectivity' is the ratio of the power returned from an object to the power you shined onto it with your laser. The reflectivity will depend on the object surface texture, material, wavelength of light etc., but it only loosely corresponds to what you perceive as bright or dark.
The metric really just reflects (:-D) the signal-to-noise ratio and dynamic range of their sensor. Max range, min/max reflectivity, SNR, accuracy, and everything else are all intertwined, so it's very difficult to compare things on equal footing unless you know exactly how it was measured. LIDAR OEMs seem to have settled on a 10%/80% rule of thumb.
My point is, I'm not sure if 20% is actually a good threshold to be measuring. What are the levels of reflectance that will be encountered in practice? I.e. if a car is black (or splattered in dirt), what percentage are we talking about? Visually a lot of cars on the road are darker than "middle gray".
Amusingly, the question of "where to set the threshold" is actually a problem for pulsed lidar in general. What constitutes the pulse returning, and how do you know which pulse is dominant? In any case, my point was that I don't think they are implying that 20% is "the one true threshold" that determines if a LIDAR is sensitive enough. Indeed, many OEMs specify the bottom bracket at 10% already, and there will always be other variables such as weather and texture that will ruin your day no matter what threshold you chose.
As with software performance, it's all about reasonable benchmarking. If you're in some lidar application, for example building a self driving car, and you currently use velodyne HDL-64 sensors that can register returns from 10% targets at 80m, then Livox's specifications give you a clue as to how their unit might compare in a similar circumstance. That's all. Past that, you have to rig up a test with the unit yourself and profile, it's the only way. I'd also add that many objects would appear different in brightness if you looked at them under a pure wavelength like 905nm, rather than the while light your eye sees.
All that said, your concerns aren't misplaced. One of the leaders in the 'new wave' lidar OEMs is Luminar, and one of their original value propositions was that they went to a different wavelength (1550nm) which has a higher eye-safe power limit. This means that they could pump out higher energy pulses, and thus get more photons back from low reflectivity targets such as tires and dark cars. The jury is still out on what works best to cover the real world range of reflectivities, largely because there are just a lot more variables at play that a simple thresholding would imply.
If this was about an aerospace application, I'm sure this sensor would have to be redundantly installed. So why do they/we assume a car needs only one of these sensors?
What makes you think a car just needs one? Many of their videos on the website suggest surrounding your car with a dozen of those sensors with overlapping fields of view.
They try to spin the scanning pattern as being an advantage as you get a more dense sampling if you point at it in a certain direction, but I'm not convinced. It is only an advantage if your device sits on a tripod. On a moving car, more predictable, structured patterns such as the Ouster OS-1 [1] make it much easier not only for deep learning, but also for SLAM (e.g. the LOAM algorithm [2] extracts feature points row by row, i.e. by ring). For a moving car or drone, any scanning pattern will sweep into a dense 3D point cloud anyway.
The pricing seems to be competitive if you only need a small field of view.
For 360 degrees, it takes four Mid-100s, which would cost $6000, or twleve Mid-40s, which would cost $7200, to obtain the same field of view coverage and the same number of points (1.2 M points per second) as a single Ouster OS-1, which is available to university researchers for $8000. However, buying a single Ouster OS-1 saves you the headache of extrinsic calibration between many lidar units, and the Ouster OS-1 only draws 14 W whereas four Mid-100s would draw a whopping 480 W. Four Mid-100s also weigh twenty times as much as one Ouster OS-1. For high density drone mapping, the Ouster OS-1 seems like a much better choice.
That said, the Livox does have a range advantage over the Ouster OS-1.
[0] (PDF) https://www.thorlabs.com/images/tabimages/Risley_Prism_Scann...
[1] https://medium.com/ouster/the-camera-is-in-the-lidar-6fcf77e...
[2] Zhang, J., & Singh, S. (2014, July). LOAM: Lidar Odometry and Mapping in Real-time. In Robotics: Science and Systems (Vol. 2, p. 9).