It's still wild to me how consumer hardware outpaced cheap lab tech so decisively around 2010-2012.

At the time I was in a research lab that needed wireless networking + high-def video + light source + on-device processing + battery power + small form factor. It turns out that even compared to Raspberry Pi, the cheapest possible solution to that problem was "buy old Androids on Ebay and add external batteries". A handful of years before that we would have been cobbling together a $500 device, or more realistically scrapping the whole project.

The whole model of feature-dense sensors suddenly crossed from something for major production runs and custom orders into single-unit consumer products, and I'm not sure people really noticed how big an impact it had on research.

I used to tear my hair at the cost of PC104 format embedded computers. No more!

This is actually a reason I am kind of surprised to see Microsoft just shut this down outright. Even as a much more limited run product, Kinect has so many possibilities in robotics and research, and it intersects with the VR/AR developments today.

I was talking to someone 3 days ago about how his company has been hoarding Kinects because they knew this day was coming.

I think robotics research is probably a reason why they're discontinuing it. Why support something that only exists outside the walled garden? Maybe not the only reason, but a reason. The device probably isn't generating profits, and the value gain on top of that is missing because the gaming community has abandoned it.

> I think robotics research is probably a reason why they're discontinuing it. Why support something that only exists outside the walled garden?

This is pretty unfair. Microsoft released Kinect 2.0 for Windows[1], with a developer SDK, specifically for the community to hack on. It even supports Windows Embedded.

I can't in my wildest dreams imagine that they thought that product would be a money maker.

[1] https://www.microsoft.com/en-us/download/details.aspx?id=445...

Similar deal as Sony removing PS3 features after launch: people found a product they loved, for reasons not envisioned by Sony, who was then hell-bent on shutting it down.

Even though it seems like Sony's interest to foster a community of high-tech nerds interested in AI and parallel processing, and even though it seems like it's Microsoft's interest in fostering a community of robotics and computer vision nerds, it turns out the Giant Corporations need lock-down control on products and have no long-term vision for supporting ecosystems or communities.

The main reason Sony killed OtherOS was that they began to think it could be used to circumvent copy protection for games. In Kinect's case I'm guessing the pricing was set at least in part with the idea that the games would make up for low-to-negative profit margins.

I'm not so sure it was so cut and dried as that. The air force was using them to create a supercomputer using other-OS and Sony was selling it as a loss.

I think they were worried about people buying them up and not purchasing games.

A professor in my college also did exactly this: https://lacal.epfl.ch/112bit_prime

They used about 200ps3 to find some prime number (cryptography related)

I think it is but for a more base reason. If they cut it, no man hours have to care for it. Simple monetary explanation. It sucked when they cut it.

There is the Asus Xtion which is a knock off Primesense Carmine. Google Tango is basically the same technology, as is Apple's Face ID*. Intel's realsense systems are also similar, but I don't know anyone who actually uses them.

Unfortunately Microsoft missed the chance to buy Primsesense before Apple snapped them up. Still, while the tech is good, the ToF sensor in the Kinect v2 is superior for the (gaming) market and Microsoft own the IP via their Canesta acquisition (they licensed the Kinect patent).

What's more interesting is the boatload of cheap ToF systems that you can now buy. The only problem is they're much more power hungry. The new Kinect is crappy for mobile robotics because its heavy (90% of the sensor head is heatsink) and needs a wall power supply. It's fine for static systems, or beefier mechanics though.

(Face ID is literally a mini-Kinect, since it comes out of Primsense IP.)

Occipital still produces the Structure Sensor [1], which uses similar technology to the 1st generation Kinect. They've also recently introduced the Structure Core [2], a new generation, miniaturized depth sensor designed to be embedded into anything that needs that kind of sensing.

[1] https://structure.io

[2] https://structure.io/core

It's not a true replacement if it costs 10-20x as much, though.

I wonder how much the face sensor on the new iPhones is going to cost? Will people be able to buy it as a separate part? I have to think that for repair purposes it will be available, but the price point will no doubt be inflated.

> Will people be able to buy it as a separate part?

Eventually, yes. I'm not sure there'll be a cloned part on Aliexpress next year, but there will be parts from broken phones for sure.

Did I miss something? It's $379 on the site. Was the kinect actually only $18.95-37.90?! Now I wish I had picked one up. I thought they were north of $100.

They are referring to the Core above - $5000 to start, $1000 for each additional sensor.

it is, if all agree that original Kinect price was subsidized based on plan to get the rest of the money through purchase of games

There are lots of alternatives today, though they took a while coming. Kinect was great, but we'll live :)

https://www.intel.ca/content/www/ca/en/architecture-and-tech...

Most of the development in realsenses has been discontinued a few months ago. We used one (SR300) at the company and discovered that some particular poses (two persons making a kissing motion) would just make the latest SDK (R3) crash, even without feature & gesture recognition active. The answer? "EWONTFIX, use the older SDK, this won't be developed anymore anyways".

Besides, the kinect has honestly better specs, a better quality in tracking and more accuracy.

We're waiting on the 400 series, which have been meant to be out "any day now" for over a year! But the SDK has gone public, and they're saying orders will open soon.

https://twitter.com/intel/status/766063343396356100

We were very disappointed with the current generation of RealSense devices. The next generation ones look good though - more based on the Tyzx acquisition.

The core is still being made for use in ther MR/VR headsets.

I'm pretty sure cheap and high quality lidar is coming very very quickly - so many companies working on it. There will be no market for an awkward kinect. Kinect is not great compared to modern lidar systems anyhow, it was just more accessible.

I thought I read Tesla got away from LIDAR because it won't be inexpensive enough for consumers any time soon.

Not entirely. LIDAR has other issues important when building self-driving cars. LIDAR can't read signs. LIDAR doesn't work well in the rain, though progress is being made these days.. And don't forget the cameras tesla uses don't rely solely on the visual spectrum we see. They can still see through fog via infrared, for example.

Can't the LIDAR eventually be made to use IR wavelengths and thus synthesize 3-d images through fog?

also, the question of seeing the fog is more nuanced that can see/can't see:

"Just like it is impossible to give a simple answer to the question “How far can I see with a thermal imaging camera?”, it is equally impossible to say how much shorter the range will be in foggy or rainy conditions. This is not only dependant on the atmospheric conditions and the type of fog but it is also dependent on the IR camera used and on the properties of the target (size, temperature difference of the target and background, etc)" [http://www.flir.com/uploadedFiles/FOG_techNote_LR.pdf]

LIDAR typically uses lasers in the so-called eye-safe range (around 1.4 micro meters). This range is precisely around the absorption peak of water so that LIDAR cannot damage the eyes of pedestrians and other by-standers. By construction, LIDAR sucks in the fog, rain, snow, etc.

But it has no problems with night. So it has some great uses.

IR LIDAR is already widely used for terrestrial mapping and meteorology. I think if that was the solution it would have already been tried.

Can different LIDARs interfere with each other? I can't see cars using it when all the cars are using it.

The timing is so precise this isn’t a huge issue.

They did, but that decision was a while ago and phased-array LIDAR has come along a lot faster than I think they expected.

Cost/unit was an incredible driver of adoption though. What other depth sensor could you pick up at a local thrift store for $10?

Maybe that was MS' problem. People that buy a 1st gen (2010) and still using it. Not upgrading it. Not replacing it. MS can't possibly profit off of it when only 1 kinect per lab is bought.

That's definitely a huge part of it. But MS would still be happy to lose money on kinect if game makers were making nice games for it.

The only game in my house that could ever utilize it was the original Dancing games from the demos. The last time I saw my daughter playing it, she was holding her phone, and using it for the sensor instead of the kinect through their integrated app.

I asked why she would do that? Just plug in the kinect and use it, don't risk dropping your phone while dancing, but that's not how she wanted to play.

Have a +1 for being a good parent and letting her play however she wanted. Gotta love kids