What's new, Atlas? (AI, Robotics and tech thread)

[Wesley_Clark]

@Snowboarder_Bo

‘Inkjet’ solar panels poised to revolutionise green energy

That’s interesting, but the bulk of solar costs are from things other than the panels themselves. Regulations, hooking them upto the grid, installation, etc. I’m not sure if that would save money.

[Darren_Garrison]

The newest in robotics. (Link may require brain-bleach.)

[Snowboarder_Bo]

You don’t even have to yell “come here, Spot!” because Spot is already coming! And he and his friends can tow their own delivery truck!

Cool little video showcasing the SpotMini as part of the marketing blitz, I suppose: the text does say that the production line is cranking them out and they will be for sale soon.

[Snowboarder_Bo]

Robot guides itself through living pig’s heart:

It is one of the first times researchers have shown that a truly autonomous surgical robot can navigate inside the heart, not controlled by a doctor with a joystick, according to a study in Wednesday’s journal Science Robotics .

Heart surgeons routinely push a thin tube called a catheter through twisting and turning blood vessels to make repairs in the heart without open surgery. But how does a robotic version find its own way through moving heart tissue and with blood swishing in the way?

Researchers at Boston Children’s Hospital turned the catheter’s camera tip into essentially an “optical whisker,” said cardiac bioengineering chief Pierre Dupont, the lead researcher. Just as cockroaches navigate along walls and rats reach out with their whiskers, the catheter maps its path through the heart, tapping periodically against the heart’s valve and wall ever so lightly — with about the force of a stick of butter sitting in your hand, Dupont said. The technology combines the camera’s images with machine learning to interpret what tissue it’s touching and how hard.

Coming soon: heart surgery by bots &/or nanobots, all done by a computer.; I’d guess within 20 years.

[Snowboarder_Bo]

Walmart is getting in on the action:

Walmart, which faces fierce competition from Amazon and other online retailers, is experimenting with digitizing its physical stores to manage them more efficiently, keep costs under control and make the shopping experience more pleasant. On Thursday, the retail giant officially opens its Intelligent Retail Lab inside a 50,000-square-foot Neighborhood Market grocery store on Long Island.

Thousands of cameras suspended from the ceiling, combined with other technology like sensors on shelves, will monitor the store in real time so workers can quickly replenish products or fix other problems.

The technology, shown first to The Associated Press, will also be able to spot spills, track when shelves need to be restocked and know when shopping carts are running low. Cameras, for example, can determine how ripe bananas are from their color, and workers will get an alert on their phone if they need to be replaced.

The article makes it clear that competition from Amazon partly prompted Walmarts investment in new tech; in the future, stores will have shoppers, but few workers, apparently.

[Snowboarder_Bo]

A company called IHMC Robotics posted a video last week showing off their LIDAR/software that helps Atlas create 3D maps of its locale and then navigate that terrain, even when only very narrow, irregular paths are available. I love how some of the footings are inadvertently unstable, yet Atlas never seems to make any obvious correction while maintaining perfect balance & composure. From the video description:

The robot senses the terrain with LIDAR and builds a map of planar regions. A path planning algorithm plans footsteps across the planar regions to a goal location, specified by an operator. The robot is currently about 50% successful over this type of terrain. We plan to increase the rate of success by adding balance using angular momentum and by better considering joint ranges of motion. Narrow terrain is difficult due to the need to do some “cross-over” steps, which are tricky due to limited range of motion in the hip joint, and also due to having a small polygon of support when one foot is directly in front of the other.

[burpo_the_wonder_mutt]

Kevin is finally gonna get his:

[Snowboarder_Bo]

I don’t even have to click; I know where that link goes (and it totally fucking rocks!).

[burpo_the_wonder_mutt]

Everybody sing:

“Bots just wanna have fu-un,
Bots just wanna have fun!”

[Snowboarder_Bo]

Behold the Translatotron!

Speak into it in one language and have your speech come out in another language!

I was alerted to this because one of my favorite YouTube channels, Two Minute Papers, just did a piece on it: All Hail The Mighty Translatotron! (6 minute video).

This seems hugely important and game-changing to me.

[Snowboarder_Bo]

So the Corridor Crew (see Post #127) offer a demonstration of Deep Fake tech where they get Tom Cruise to come to their studio. Frankly, this is kinda scary. Fascinating, but scary.

They’re wrong that that there’s no Deep Fake for voices tho; we’ve covered several companies handling that earlier in the thread and things have only gotten better in the months since then. That link goes to a news article that has an embedded video with a CG voice that mimics Joe Rogan. Now, I don’t listen to him but I know he’s a sports comedy guy and I’m certain I’ve heard him before… and I can’t at all tell that this 100% computer generated voice isn’t a real person talking. I’d be really interested in having a Joe Rogan fan listen to it.

I think what’s really scary about this Deep Fake video is knowing that it was done on a machine that is affordable to many, many people AND that it took so little time to train the AI. IMO 2-3 days isn’t a lot of time. Imagine letting it train itself for 100 days with 10000x the reference images, and it seems to me that this kind of thing is going to quickly become indistinguishable from real recorded footage.

---

Oh, and now there’s AIs being trained to produce a controllable full-body image. In other words, they can now digitize a person and a joystick/controller can be used to manipulate the image in real time. The AI figures out how to get from one body pose to another realistically. So soon you won’t need a whole body suit to do animations. Hollywood can just have 2 people in comfortable studio gaming chairs make AI CGI Jackie Chan fight AI CGI Arnold Schwarzenegger as tho it were a video game and then edit all the really cool parts into a fight scene.

I’m now VERY sure that my lifetime is gonna include some VERY fucked up shit happening.

[Darren_Garrison]

@Snowboarder_Bo

.I’m now VERY sure that my lifetime is gonna include some VERY fucked up shit happening.

What a time to be alive! (Sniff.)

Almost makes up for the no flying cars.

[ftg]

Just want to mention that a new set of Raspberry Pi’s have been announced. The Pi 4 comes in 1/2/4 GB versions (at $35/$45/$55 resp.).

Newer, much faster, quad-core processor. USB 3.0 ports. True gigabit Ethernet. Dual (micro) HDMI ports. 4k resolution (supposedly). New GPIO options. Etc.

Don’t need the dual screens myself and I hate micro HDMI. Just another adapter to deal with. Also the power port is USB C now (and it draws more power, of course). Not case compatible. (So an upgrade requires several new parts.)

Tom’s Hardware has a review. Note the immense speed increase when using a SSD. This will make a big difference for me.

Wonder what new AI stuff you can do with it.

[Darren_Garrison]

This bit on watermarks is from a year ago, but I just saw it today. (It would be great if you could use this to do on the fly de-buging of video as you watch.)

[N9IWP]

Insect like robot close to independent flight:

Flying, insect-like robot flits closer to independent flight

No control, and for less than a second, but still cool IMHO

Brian

[Darren_Garrison]

Something that really interests me is that we are beginning to see LIDAR on cell phones (under the name TOF or Time Of Flight sensor.) The main purpose is to be able detect depth for focusing images, but there are other, much cooler potentials. Such as using your phone camera to create 3D models of objects (and people) you photograph, and measuring dimensions and volumes of objects. The sensors are on only a few high-end phones now, but they will trickle down.

(There will probably also be cool secondary uses for the up-and-coming short-range RADAR.)

[Darren_Garrison]

How about a creepy new iPhone feature?

[Snowboarder_Bo]

@Darren_Garrison

Something that really interests me is that we are beginning to see LIDAR on cell phones (under the name TOF or Time Of Flight sensor.) The main purpose is to be able detect depth for focusing images, but there are other, much cooler potentials. Such as using your phone camera to create 3D models of objects (and people) you photograph, and measuring dimensions and volumes of objects. The sensors are on only a few high-end phones now, but they will trickle down.

And don’t forget that every time someone uses it, it trains the AI better. It’s going to get very good very quickly with hundreds of millions of people using it, even if 90% only use it once.

@Darren_Garrison

(There will probably also be cool secondary uses for the up-and-coming short-range RADAR.)

This sounds like really, really cool tech. If it works simply and works well, it’s prolly game changing. Video gaming would change overnight. A good portion of the machines in Las Vegas would be changed within a year; no more slapping buttons or pulling levers. TV sets, computers, clocks, doors, elevators, etc., etc.

[Snowboarder_Bo]

Check this out: This AI Makes The Mona Lisa Come To Life.

This AI can create video of a person’s face moving, saying things, changing expressions, etc. from a single image. Any face. Even a painted one. It is amazing.

CGI movies are gonna be amazing in 10 years.

[Snowboarder_Bo]

More than 3 million people work in the US as migrant and seasonal farmworkers; they are all going to be out of job in the next 10 years: Robot uses photonic sensors to pick strawberries in gee-whiz numbers.

Human pickers can collect around 50 kilograms in a day but take breaks.

The Scottish Farmer walked readers through the impressive numbers of yield and labor output that would result in using the robots, "promising a ‘revolution’ in harvesting soft fruit, with a workrate of 11,500 berries in a 16-hour day.

The description said they were bruise-free, every 5 seconds, with the robot’s delicate clasping mechanism, and with the ability to deliver between 180 and 360 kilograms every single day.

Jon Excell, The Engineer named two key features of this robot— smart photonics and a patented soft-touch gripper. The robot can both pick strawberries and sort them by size or weight and pack as it goes along. The strawberries are picked based on ripeness and size. They are packaged in varied categories.

The Scottish Farmer said the machine, and all within a 5-second picking cycle, (1) analyzes colors of each fruit and (2) assesses if the fruit is ripe enough for picking via the clasping mechanism.

The article even addresses the human worker aspect of the machine’s use:

So, where does that leave human workers? Gordon Davidson quoted Jan Anthonis of the company, who said, “Rotting and unpicked fruit from a lack of human pickers on farms all over the world could soon be tackled with robots.”

“You no longer need to worry about finding the right people but you can focus on tomorrow,” according to the company web site.

So yeah: all those people are about to be unemployable in farmwork. Might take 5 or 10 or 20 years, but there’s simply no way that farms aren’t going to utilize this technology. They do 3.5x to 7+x the work that a person does, with less waste and zero downtime. Hell, people could pick all day and night and still not equal what the robot can pick in 8 hours. That’s a no brainer for anyone using human labor.