Command Line Heroes - Robot as Servant

Episode Date: September 7, 2021

The 1980s promised robotic servants were in reach. They’d clean up our houses. Bring us drinks. Usher in an era of leisure. We didn’t get robot butlers. But if we look around, we’ll find an army of robotic servants already automating away domestic drudgery. Richard Rowland recounts the extent to which Androbot over-promised on its ability to build a robot servant. 40 years later, we still don’t have robot maids. Monroe Kennedy III walks us through the complexities of seemingly simple tasks. To make things more difficult, each attempt to build a robot had to build the hardware AND write the code from scratch. Keenan Wyrobek explains that’s why he helped write and share the Robot Operating System (ROS). Leila Takayama describes how beneficial ROS was to the field of robotics. And Terry Fong shares how NASA is using ROS to build the robots that explore our solar system.If you want to read up on some of our research on domestic robots, you can check out all our bonus material over at redhat.com/commandlineheroes. Follow along with the episode transcript.

Where are my slippers? Right here, sir. Thank you, dear. You're welcome, sir. Hey, get me a beer, would you? My pleasure. Did you ever watch The Jetsons? A very mid-century view of the future.

And each episode, at some point, Rosie the robot would wheel in and help the family out with some domestic chore. She might make a wisecrack, but basically, Rosie was an obliging artificial intelligence whose whole purpose in life was to wipe down windows, vacuum the carpet, and fetch your slippers. That was the fantasy. A pretty sexist fantasy. And it was our idea of what a robot could be. A domestic servant. A maid that was made of metal. I'm Saranya Barg, and this is Season 8 of Command Line Heroes, an original podcast from Red Hat.

This season, we're tackling one question from every angle. What is a robot? Since the word was coined almost exactly a century ago, the idea of robots has fascinated us and sometimes terrified us. Science fiction gave us robots that help and robots that rebel. In our novels and movies, robots embody every hope and fear we've got about technology. People rode monorails to world fairs where the 21st century was laid out as one big bonanza of convenience and high-tech magic. But now that our robot future is coming to life, what are they really like? What have we actually made? And what might they become?

While making this season, we discovered a robot reality that's pretty removed from the robots we imagined. And yet, all those misguided ideas about what a robot should be actually shaped what robots are. And no robot fiction shaped things more than the story we told ourselves about robotic servants. We thought they'd become our butlers and maids, machines like Rosie on the Jetsons. That idea reached a fever pitch in the mid-80s, maybe because our tech had evolved just enough so we could produce, you know, a calculator on wheels. We got a little excited. For the first time, we could build the image of the robots we'd been imagining. And that was enough to make us think, any day now, robots will be part of ordinary domestic life.

In 1983, it was very believable that we would have personal robots. Rick Rowland lives in Las Vegas, where he manages an RV dealership. But Rowland has a side hustle. He builds robots. And he's become a kind of robot historian, too. He's lost count, but there's upwards of 200 specimens in his collection. Robots from an earlier time, when each new model seemed to herald a robot revolution. His robot collection, full of dusty wheels and plastic faces, reminds us that, back in the 80s, expectations were high. The future was fantastic. In my lifetime, I had gone from no cell phones, no computers, no nothing. And then when I started college, they were using the old-fashioned mainframe

where you don't even have a screen, it's just a paper coming out of the machine. And just in those few years, we went to buying a home computer over at Kmart or Jimco, and they did amazing things. So a robot should be able to just follow along since a robot is basically a computer on wheels. Every year, Rowan would visit the big computer electronics show that came to Vegas. In the 80s, conventions like that were stoking the public's desire for robots.

You could cruise down the aisles and find all these hopeful, but usually dead-end, inventions. Every new robot company was promising a consumer product that could roll around your living room, serving canapes and dusting the chandelier. Unfortunately, the marketing department worked a lot harder at it than the production department. The company with some of the most enticing promises was called Androbot. It was founded by Nolan Bushnell. You might recognize that name. He's the same guy who founded Atari. And his robots were more like video games than anything else. Bushnell produced a machine called Bob, which stood for brains on board. And there was another model

called Topo. These were boxy R2-D2 style robots whose real purpose was to excite the crowds. Our robot expert doesn't let this history die away. He saved the pamphlets he collected at those old CES shows, and we got him to read read us from an Androbot pamphlet from 1983. I grabbed a bunch of them just in case I had to get a memory jog. The year is 1 A.B. And of course, the A.B. was for Androbot. Open a new era in technology history, a dream has engaged man's imagination for centuries.

The microprocessor now makes possible the personal robot and a profound change in all our lives about to take place. And what kind of change, you ask? Bushnell's salesmen knew exactly what their customers were looking for. The first robot I remember seeing at CES in 1983 was the Androbot booth, and inside that booth was a robot called Bob. He said, Bob, go get me a beer. And he turned around, truddled over to the refrigerator. Of course, it was a custom-built refrigerator that dispensed beer out of the side. It dropped into a special holder they had rigged up on the side of Bob, and it came back over to him, and he opened the beer and says, thank you, Bob. So that was pretty amazing. Everybody, especially in the 80s, wanted robots to bring them a beer.

If a robot could bring a dude a beer, it had basically just passed the Turing test as far as these guys were concerned. But beyond the fake waitress act, there were a bunch of smaller, more practical jobs that were actually within a robot's reach. Robots were going to detect fires or roam your hallway at night to scare away intruders. They might answer your phone and take messages too. And of course, a lot of that technology has ended up in our pocket. So unfortunately, all of the robotic ideas of the 80s disappeared as the technology got smaller and more advanced. Basically, a lot of the tasks that robots could handle ended up being miniaturized into smoke detectors and smartphones. But what about all the stuff that actually requires a body?

Can the robot do laundry? They always wanted to do domestic chores. Those jobs remained stubbornly out of reach, because it turns out, folding laundry, putting away the groceries, everything these robot designers thought of as menial labor wasn't nearly as simple as they imagined. Bushnell and others had promised big things, a Rosie the Robot made in every home. But they didn't have the first idea how hard it is to simply be a robot navigating our big, messy world. Today, we've managed to fly robot helicopters on Mars. Not bad. But how about robots that can grab me a drink?

It turns out that supposedly simple actions can be incredibly complex. The robot as domestic servant is a far more ambitious dream than we ever knew. Now, how do you handle the generic beer that's not always placed in the right position? Monroe Kennedy is an assistant professor at Stanford. He directs their assistive robotics and manipulation laboratory. When we asked Kennedy about those beer-fetching robots that folk like Nolan Bushnell were trying to sell, he pointed out that those robots could never have fetched a beer in anybody's actual kitchen.

You always place the beer in the same spot. It's going to do something cool with it, but everything has to be just so, right? If you were to change the position of one of the machines by six centimeters, the whole process might break down. To build a robot that can navigate the real world in a meaningful way, the real world with all its inconsistent elements, that's an enormous challenge. For example, 40 years after those Androbot models were strutting around at trade shows, Kennedy is still working to build a robot that can pick up a glass of water. We asked what was taking so long, and the answer

was not short. Yeah, let's assume that this glass of water is literally in a transparent glass. The first problem that you'd have to address is perception. How does the robot see the glass? Okay, hadn't thought about that. Gonna be hard to know where the boundary of a transparent surface is. You have to leverage basic knowledge about light distortion, stuff like that. A lot of these sensors that we use all the time, be it in LIDARs in cars or Kinects if you're playing, you know, the PlayStation, right? All of these break down very quickly when you think about a transparent object. And so all of a sudden, there's a whole class of sensors you can't use from, you know,

vision or perception at a distance. And so perhaps you need stereo cameras. If we go back to this notion over where this feature detection has come over the decades, usually it's the detection of lines, corners, edges, dots. But then if I look at a glass that's been distorted, I don't really always have nice edges, corners, or dots to detect. And so there are no good features from all of the past decades of work that we use to detect other things. The problems just keep piling up. Now you go to grab this object, but maybe the glass is sweating. Maybe it's wet.

So a novice robot would grasp it, say, hey, it's in my hands. I've squeezed it. I go to pick it up. It slips out of my hand and breaks. So now this means I need new hardware in my fingers such that as I grab it, I can detect not only the normal force, but also the friction. As you're moving this glass, can you detect how much water is inside? If you move it in motion that is too jerky, this could cause the water to spill and damage the electronics on you or in the area. There's more. An endless list of problems. He hasn't even gotten to carrying the glass across the room. The basic takeaway is picking up a glass

of water feels next to impossible. Still, we're not ready to quit the robot game. And it's good we haven't given up because a recent shift in the robotics ecosystem could change everything. Billionaire Scott Hansen wanted to accelerate the evolution of robotics. So, in 2006, he started funding a research lab called Willow Garage. Unlike earlier institutions, which tried to produce luxury consumer robots, Willow Garage would be closer to pure science. The idea was to push the whole field of robotics forward for everybody. By 2010, the folks at Willow Garage had developed a robot of their own, the PR2. It's pretty cool looking, vaguely humanoid if you squint your eyes.

And the PR-2 can do some amazing tricks. Domestic chores like folding your laundry, maybe grab you that beer from the fridge. At first glance, you might think, okay, so they're a couple steps closer to building a robotic servant. But those tricks, as cool as they are, they aren't the point of the work at Willow Garage. In fact, the real robotic breakthrough lies under the PR2's metallic skin, beneath that old idea of a metal maid. To understand what I mean, you need to know about Kenan Weirobeck and Eric Berger, two young Stanford graduates who were among the first employees at Willow.

They had a mission. I always wanted to have those better tools and be able to make better tools for everybody. Today, Weirobeck is the founder and CTO of Zipline, where they create drone delivery systems for medical supplies. But back in the early 2000s, he was frustrated by the roadblocks in robotics. Everybody in the field seemed to spend all their time reinventing the wheel. There's so many disciplines required to make a robotic system useful. And each one of them is hard, like deeply hard, like hard math and complex software.

Imagine you wanted to design a better kind of house. But first, you have to invent basic insulation and electrical wiring and plumbing. Oh, and hammers. This is nuts. A typical PhD in robotics was, you know, four years or so of basically writing software, integrating hardware that has been done before. Very little innovation in that space. There had to be a better way. Weirbeck and Berger imagined a robotics revolution that would open source their field in the same way Linux had done for the world of software. Linux was clearly enabling the internet, unlike any closed computer operating system was. And because you can enable people to build all these different pieces and other people to put all the pieces together and

actually create useful things really easily. That was like, okay, robotics is missing this, we're gonna make this. And when Weyerbeck and Berger arrived at Willow Garage, suddenly they had the time and resources to do exactly that. What we called back then the Linux of robotics eventually came up with a really creative name called Robot Operating System. This would be the secret weapon inside the PR2 robot,

a new kind of robotic operating system. Keep in mind, unlike the proprietary attitudes at those robot conventions in the 80s, Weyerbeck and Berger's work was a fully funded open source effort. They had the chance to create a set of tools that everyone could use. Ross, the robot operating system,

would give future roboticists two major gifts. First, middleware. Middleware enables you to basically plug different software libraries for doing different types of robotic math together with different sensors, with different actuators on different computers. And then on top of it, it's a bunch of tools that basically make, that leverage that common middleware standard to do things like visualize your data, log your data, replay your data, really debug your system.

And all that was just for starters. Being open source, the real magic comes when the community arrives. The ROS community started building libraries that work with all that middleware. One library might help you with your robot's arm movements. Another might let your robot guide its way around a room or detect a human being. Imagine a whole bank of prefab skills you could draw from. ROS makes all that sharing possible. But software alone is never enough.

If you wanted to use ROS in the early days, even if you were a very smart person, it was still hard. Layla Takayama, a human-robot interaction researcher based in Santa Cruz, was among the first to discover ROS. In 2009, she came to Willow Garage, where she brought a human-centered perspective that was sometimes missing among all the robot fanatics. Making ROS more accessible and user-friendly was crucial. We had aspirations for ROS to be bigger. We wanted all of the major universities and research institutions and companies to use it. Problem was, ROS hadn't had time to evolve. Compared to something like Linux, which had grown up over decades, the robot operating system was still in its infancy and pretty confusing to newcomers.

Everybody struggled. Every single grad student, no matter how smart they were, had a hard time. Our first reaction was like, oh my gosh, we need to go and help them. And so, you know, the software team would go in there and hold their hand. And that was not good because that's exactly the thing that doesn't scale. Takayama would insist that user studies be run like real-world scenarios. She also pushed for greater documentation. And slowly, year by year,

the ROS community gathered its own momentum. It became coherent, letting more and more people join in. A lot of companies that are using ROS today to build real products and services are helping to make that a higher priority. Or they're doing it in-house, and some of them are even releasing code back, which is great. Meanwhile, the folks at Willow Garage could use their PR2 robot to show off the magic of ROS. PR2 became a kind of platform robot that let them exercise their code.

To the outside world, it might have looked like one more Rosie the maid, another fun toy to boss around. But to the Willow Garage team, PR2 and its ROS brains were so much more. They were a way to level up robotics everywhere. In 2010, Willow even sent the PR2 out into the world, donating $4 million worth of robots to a dozen research facilities, sort of like robot evangelists for the gospel of Ross. And all this is starting to work. As of 2012, Ross is thriving at a new organization called Open Robotics that's devoted to its development. It feels like we finally got that common toolkit that everybody can use. Ross managed to build enough critical mass and enough positive energy

and a good culture that actually engaged enough people to play together and to do it in a more open way. And I think that is what can create a revolution. It's a revolution where anybody can leap into action because nobody has to start from scratch. What ROS has enabled people to do is you can form a company that's focused on an actual application with actual users who have a problem to solve.

And you can leverage the entire world of robotics, you know, robotics community just by using ROS. Suddenly, the robots we thought we were building, the robots that would clean our houses and fetch our drinks, could stand on the shoulders of robot giants. They could reach higher every generation. And today, they're even reaching for the stars. Okay, this is Terry Fong. I am the chief roboticist at the NASA Ames Research Center in California. With open-source robotics, the sky is not the limit. Even NASA benefits. At the International Space Station, for example, our dreams of domestic robots will soon level up with flying robots. This project, called Astrobee, handles housekeeping duties in zero gravity. Autonomous, free-flying, robotic crewmates for the astronauts. And these new helpers are being developed

thanks to open-source software. One of the decisions we made early on when we were developing Astrobee was to make it easily accessible so that the guest scientists could work with it. So we chose to release all of the software for controlling the robots, all the software tools. For example, we have a pretty high-fidelity 3D dynamic simulator so people can simulate as they're doing development how Astrobee will fly around. All of that we released as open source and we put the

source code out on GitHub. It doesn't stop there either. In 2023, the Viper robot will be searching the south pole of the moon for frozen water. And thanks to open source innovations, computers on Earth will be able to send

Viper instructions. Part of the software we're using to do that on the ground is ROS. Fong feels that larger open source communities have actually increased the reliability of the robotics he works with. If it's something that's adopted by a community, it becomes sustainable. And it's something that people continually work to improve, which means the quality goes up. But more than that, it's something that you can count on being there because people care about it. Together, we've come a long way from Rosie the Mane. Our goals for robots have changed wildly in the past few decades. What we imagined as an endpoint, those domestic robot servants, turned out to be just the starting point for a robot revolution that's rocketing to other worlds. And for Fong, that means we can form a new kind of relationship

with our robots. We don't have to limit our imaginations, trying to make them into elaborate house cleaners and personal assistants. As we send humans deeper into space, what's needed to make them successful? Going to Mars will be unlike anything we've ever done before. And how do you make sure that, say, a habitat where humans are going to live, is operational and it's safe and it's secure so that when they get there, they don't have to scramble to just stay alive. Those of us who work in space robotics fully believe that robots should be part of the solution.

A fantasy that, half a century ago, involved Rosie the robot cleaning up your living room has evolved into something so much more. But all that work on basic chores for robots did push the industry forward. And soon, Rosie the Robot will be fixing up our home on Mars. You know, even today, there's still no robot butler, no robot maid.

But maybe that's okay. Maybe you can get your own beer. The robot dream of the 1980s lives on anyway as a kind of museum collection over at Rick Rowland's place. I've collected anything made by Androbot. The Topo series, the Bob series, the Fred series. There was Andy. Oh, have the Fred series. There was Andy. Have the hero line.

When you look at all those dusty old models, you realize it was the dream that mattered. By dreaming up robots that would help us with ordinary tasks, we pushed ourselves, inspired ourselves, to create robots that would one day help with the extraordinary. Ross and the emergence of open-source robotics created the potential for a robot revolution far beyond the gussied-up Roombas of the 80s. It's a software revolution,

and that's what we're exploring next time in Episode 2. I'm Saranya Dbarik, and this is Command Line Heroes, an original podcast from Red Hat follow or subscribe wherever you get your podcasts and until next time keep on coding the Hero Junior we have the RB5X robot

and the RB5X robot with arms a robot that came along a little later was called the Newton Sinpet we have all the Omnibot series from Tommy. We have Robbie Jr.'s, Robbie Sr.'s. Hi, I'm Mike Ferris, Chief Strategy Officer at Red Hat. And as you might expect in my role, I get a lot of questions about AI, particularly about foundation models. Now, don't get me wrong.

Those are important, but they're not the whole story. Whether you're using a commercial model or an open source one, you're going to need to fine tune or augment models with your data for your use case. And you need a common platform for that, where data scientists, app developers, and ops teams can all collaborate, especially as you start to scale. And then this is iterative. It's rinse and repeat. So really, it's about making that fast path from idea to model to production and back again. And that's what Red Hat

OpenShift AI does. Head to redhat.com to learn more.