When IEEE Spectrum first wrote about Covariant in 2020, it was a new-ish robotics startup looking to apply robotics to warehouse picking at scale through the magic of a single end-to-end neural network. At the time, Covariant was focused on this picking use case, because it represents an application that could provide immediate value—warehouse companies pay Covariant for its robots to pick items in their warehouses. But for Covariant, the exciting part was that picking items in warehouses has, over the last four years, yielded a massive amount of real-world manipulation data—and you can probably guess where this is going.

Today, Covariant is announcing RFM-1, which the company describes as a robotics foundation model that gives robots the “human-like ability to reason.” That’s from the press release, and while I wouldn’t necessarily read too much into “human-like” or “reason,” what Covariant has going on here is pretty cool.

“Foundation model” means that RFM-1 can be trained on more data to do more things—at the moment, it’s all about warehouse manipulation because that’s what it’s been trained on, but its capabilities can be expanded by feeding it more data. “Our existing system is already good enough to do very fast, very variable pick and place,” says Covariant co-founder Pieter Abbeel. “But we’re now taking it quite a bit further. Any task, any embodiment—that’s the long-term vision. Robotics foundation models powering billions of robots across the world.” From the sound of things, Covariant’s business of deploying a large fleet of warehouse automation robots was the fastest way for them to collect the tens of millions of trajectories (how a robot moves during a task) that they needed to train the 8 billion parameter RFM-1 model.

Covariant

“The only way you can do what we’re doing is by having robots deployed in the world collecting a ton of data,” says Abbeel. “Which is what allows us to train a robotics foundation model that’s uniquely capable.”

There have been other attempts at this sort of thing: The RTX project is one recent example. But while RT-X depends on research labs sharing what data they have to create a dataset that’s large enough to be useful, Covariant is doing it alone, thanks to its fleet of warehouse robots. “RT-X is about a million trajectories of data,” Abbeel says, “but we’re able to surpass it because we’re getting a million trajectories every few weeks.”

“By building a valuable picking robot that’s deployed across 15 countries with dozens of customers, we essentially have a data collection machine.” —Pieter Abbeel, Covariant

You can think of the current execution of RFM-1 as a prediction engine for suction-based object manipulation in warehouse environments. The model incorporates still images, video, joint angles, force reading, suction cup strength—everything involved in the kind of robotic manipulation that Covariant does. All of these things are interconnected within RFM-1, which means that you can put any of those things into one end of RFM-1, and out of the other end of the model will come a prediction. That prediction can be in the form of an image, a video, or a series of commands for a robot.

What’s important to understand about all of this is that RFM-1 isn’t restricted to picking only things it’s seen before, or only working on robots it has direct experience with. This is what’s nice about foundation models—they can generalize within the domain of their training data, and it’s how Covariant has been able to scale their business as successfully as they have, by not having to retrain for every new picking robot or every new item. What’s counter-intuitive about these large models is that they’re actually better at dealing with new situations than models that are trained specifically for those situations.

For example, let’s say you want to train a model to drive a car on a highway. The question, Abbeel says, is whether it would be worth your time to train on other kinds of driving anyway. The answer is yes, because highway driving is sometimes not highway driving. There will be accidents or rush hour traffic that will require you to drive differently. If you’ve also trained on driving on city streets, you’re effectively training on highway edge cases, which will come in handy at some point and improve performance overall. With RFM-1, it’s the same idea: Training on lots of different kinds of manipulation—different robots, different objects, and so on—means that any single kind of manipulation will be that much more capable.

In the context of generalization, Covariant talks about RFM-1’s ability to “understand” its environment. This can be a tricky word with AI, but what’s relevant is to ground the meaning of “understand” in what RFM-1 is capable of. For example, you don’t need to understand physics to be able to catch a baseball, you just need to have a lot of experience catching baseballs, and that’s where RFM-1 is at. You could also reason out how to catch a baseball with no experience but an understanding of physics, and RFM-1 is not doing this, which is why I hesitate to use the word “understand” in this context.

But this brings us to another interesting capability of RFM-1: it operates as a very effective, if constrained, simulation tool. As a prediction engine that outputs video, you can ask it to generate what the next couple seconds of an action sequence will look like, and it’ll give you a result that’s both realistic and accurate, being grounded in all of its data. The key here is that RFM-1 can effectively simulate objects that are challenging to simulate traditionally, like floppy things.

Covariant’s Abbeel explains that the “world model” that RFM-1 bases its predictions on is effectively a learned physics engine. “Building physics engines turns out to be a very daunting task to really cover every possible thing that can happen in the world,” Abbeel says. “Once you get complicated scenarios, it becomes very inaccurate, very quickly, because people have to make all kinds of approximations to make the physics engine run on a computer. We’re just doing the large-scale data version of this with a world model, and it’s showing really good results.”

Abbeel gives an example of asking a robot to simulate (or predict) what would happen if a cylinder is placed vertically on a conveyor belt. The prediction accurately shows the cylinder falling over and rolling when the belt starts to move—not because the cylinder is being simulated, but because RFM-1 has seen a lot of things being placed on a lot of conveyor belts.

“Five years from now, it’s not unlikely that what we are building here will be the only type of simulator anyone will ever use.” —Pieter Abbeel, Covariant

This only works if there’s the right kind of data for RFM-1 to train on, so unlike most simulation environments, it can’t currently generalize to completely new objects or situations. But Abbeel believes that with enough data, useful world simulation will be possible. “Five years from now, it’s not unlikely that what we are building here will be the only type of simulator anyone will ever use. It’s a more capable simulator than one built from the ground up with collision checking and finite elements and all that stuff. All those things are so hard to build into your physics engine in any kind of way, not to mention the renderer to make things look like they look in the real world—in some sense, we’re taking a shortcut.”

RFM-1 also incorporates language data to be able to communicate more effectively with humans. Covariant

For Covariant to expand the capabilities of RFM-1 towards that long-term vision of foundation models powering “billions of robots across the world,” the next step is to feed it more data from a wider variety of robots doing a wider variety of tasks. “We’ve built essentially a data ingestion engine,” Abbeel says. “If you’re willing to give us data of a different type, we’ll ingest that too.”

“We have a lot of confidence that this kind of model could power all kinds of robots—maybe with more data for the types of robots and types of situations it could be used in.” —Pieter Abbeel, Covariant

One way or another, that path is going to involve a heck of a lot of data, and it’s going to be data that Covariant is not currently collecting with its own fleet of warehouse manipulation robots. So if you’re, say, a humanoid robotics company, what’s your incentive to share all the data you’ve been collecting with Covariant? “The pitch is that we’ll help them get to the real world,” Covariant co-founder Peter Chen says. “I don’t think there are really that many companies that have AI to make their robots truly autonomous in a production environment. If they want AI that’s robust and powerful and can actually help them enter the real world, we are really their best bet.”

Covariant’s core argument here is that while it’s certainly possible for every robotics company to train up their own models individually, the performance—for anybody trying to do manipulation, at least—would be not nearly as good as using a model that incorporates all of the manipulation data that Covariant already has within RFM-1. “It has always been our long term plan to be a robotics foundation model company,” says Chen. “There was just not sufficient data and compute and algorithms to get to this point—but building a universal AI platform for robots, that’s what Covariant has been about from the very beginning.”

The global ocean is difficult to explore—the common refrain is that we know less about the deep ocean than we do about the surface of the moon. Australian company Advanced Navigation wants to change that with a pint-sized autonomous underwater vehicle (AUV) that it hopes will become the maritime equivalent of a consumer drone. And the new AUV is already getting to work mapping and monitoring Australia’s coral reefs and diving for shipwrecks.

The Sydney-based company has been developing underwater navigation technology for more than a decade. In 2022, Advanced Navigation unveiled its first in-house AUV, called Hydrus. At less than half a meter long, the vehicle is considerably smaller than most alternatives. Even so, it’s fully autonomous and carries a 4k-resolution camera capable of 60 frames per second that can both capture high-definition video and construct detailed 3D photogrammetry models.

Advanced Navigation says Hydrus—with a depth rating of 3,000 meters, a range of 9 kilometers, and a battery that lasts up to three hours—is capable of a wide variety of missions. The company recently sold two units to the Australian Institute of Marine Science (AIMS), the country’s tropical marine science agency, which will use them to survey coral reefs in the North West Shelf region off the country’s west coast. Hydrus has also recently collaborated with the Western Australian Museum to produce a detailed 3D model of a shipwreck off the coast near Perth.

“If people can go and throw one of these off the boat, just like they can throw a drone up in the air, that will obviously benefit the exploration of the sea.” —Ross Anderson, Western Australian Museum

After many years of supplying components to other robotics companies, Peter Baker, subsea product manager at Advanced Navigation, says they company spotted a gap in the market. “We wanted to take the user experience that someone would have with an aerial drone and bring that underwater,” he says. “It’s very expensive to get images and data of the seabed. So by being able to miniaturize this system, and have it drastically simplified from the user’s point of view, it makes data a lot more accessible to people.”

But building a compact and low-cost AUV is not simple. The deep ocean is not a friendly place for electronics, says Baker, due to a combination of high pressure and corrosive seawater. The traditional way of dealing with this is to stick all the critical components in a sealed titanium tube that can maintain ambient pressure and keep moisture out. However, this requires you to add buoyancy to compensate for the extra weight, says Baker, which increases the bulk of the vehicle. That means bigger motors and bigger batteries. “The whole thing spirals up and up until you’ve got something the size of a minibus,” he says.

Advanced Navigation got around the spiral by designing bespoke pressure-tolerant electronics. They built all of their circuit boards from scratch, carefully selecting components that had been tested to destruction in a hydrostatic pressure chamber. These were then encapsulated in a water-proof composite shell, and to further reduce the risk of water ingress the drone operates completely wirelessly. Batteries are recharged using inductive charging and data transfer is either over Wi-Fi when above water or via an optical modem when below the surface.

Hydrus AUVs are charged using induction to keep corrosive seawater from leaking in through charging ports.Advanced Navigation

This has allowed the company to significantly miniaturize the system, says Baker, which has a drastic impact on the overall cost of operations. “You don’t need a crane or a winch or anything like that to recover the vehicle, you can pick it up with a fishing net,” he says. “You can get away with using a much smaller boat, and the rule of thumb in the industry is if you double the size of your boat, you quadruple the cost.”

Just as important, though, is the vehicle’s ease of use. Most underwater robotics systems still operate with a tether, says Baker, but Hydrus carries all the hardware required to support autonomous navigation on board. The company’s “bread and butter” is inertial navigation technology, which uses accelerometers and gyroscopes to track the vehicle from a known starting point. But the drone also features a sonar system that allows it to stay a set distance from the seabed and also judge its speed by measuring the Doppler shift on echoes as they bounce back.

This means that users can simply program in a set of way points on a map, toss the vehicle overboard and leave it to its own devices, says Baker. The Hydrus does have a low-bandwidth acoustic communication channel that allows the operator to send basic commands like “stop” or “come home,” he says, but Hydrus is designed to be a set-and-forget AUV. “That really lowers the thresholds of what a user needs to be able to operate it,” he says. “If you can fly a DJI drone you could fly a Hydrus.”

The company estimates for a typical seabed investigation in water shallow enough for human divers, the Hydrus could be 75 percent cheaper than alternatives. And the savings would go up significantly at greater depths. What’s more, says Baker, the drone’s precise navigation means it can produce much more consistent and repeatable data.

To demonstrate its capabilities, Hydrus’ designers went hunting for shipwrecks in the Rottnest ships graveyard just off the coast near Perth, in Western Australia. The site was a designated spot for scuttling aging ships, says Ross Anderson, curator at Western Australian Museum, but has yet to be fully explored due to the depth of many of the wrecks.

The Advanced Navigation team used the Hydrus to create a detailed 3D model of a sunken “coal hulk”—one of a category of old iron sailing ships that were later converted to floating coal warehouses for steamships. The Western Australian Museum has been unable to identify the vessel so far, but Anderson says these kind of models can be hugely beneficial for carrying out maritime archaeology research, as well as educating people about what’s below the waves.

Advanced Navigation used its new Hydrus drone to create a detailed 3D image of an unidentified “coal hulk” ship in the Rottnest ships graveyard off the western coast of Australia.

Advanced Navigation

Any technology that can simplify the process is greatly welcomed, Anderson adds. “If people can go and throw one of these off the boat, just like they can throw a drone up in the air, that will obviously benefit the exploration of the sea,” he says.

Ease of use was also a big driver behind AIMS’s purchase of two Hydrus drones, says technology development program lead Melanie Olsen, who is also an IEEE senior member. Most of the technology available for marine science is still research-grade and a long way from a polished, professional product.

“When you’re an operational agency like AIMS, you typically don’t have the luxury of spending a lot of time on the back of the boat getting equipment ready,” says Olsen. “You need something that users can turn on and go and it’s just working, as time is of the essence.”

Another benefit of the Hydrus for AIMS is that the drone can operate at greater depths than divers and in conditions that would be dangerous for humans. “Its enabling our researchers to see further down in the water and also operate in more dangerous situations such as at night, or in the presence of threats such as crocodiles or sharks, places where we just wouldn’t be able to collect that data,” says Olsen.

The agency will initially use the drones to survey reefs on Australia’s North West Shelf, including Scott Reef and Ashmore Reef. The goal is to collect regular data data on coral health to monitor the state of the reefs, investigate how they’re being effected by climate change, and hopefully get early warning of emerging problems. But Olsen says they expect that the Hydrus will become standard part of their ocean monitoring toolkit going forward.

This story was updated on 11 March 2024 to correct the year when Advanced Navigation unveiled Hydrus.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

[ CMU ]

[ Takahiro Miki ]

Sure, 3.3 meters per second is fast for a humanoid, but I’m more impressed by the spinning around while walking downstairs.

[ Unitree ]

[ SAQIEL ]

Thanks, Temma!

[ Robot Era ]

“Impact-Aware Bimanual Catching of Large-Momentum Objects.” Need I say more?

[ SLMC ]

[ UMass Amherst ]

Thanks, Julia!

The manipulation here is pretty impressive, but it’s hard to know how impressive without also knowing how much the video was sped up.

[ Somatic ]

[ DJI ]

[ Extend Robotics ]

[ Cybathlon ]

It’s been a long road to electrically powered robots.

[ ABB ]

[ CRASAR ]

It probably will not surprise you that there are a lot of robots involved in building Rivian trucks and vans.

[ Kawasaki Robotics ]

[ DARPA ]

You’ve seen this before: a truck unloading robot that’s made up of a mobile base with an arm on it that drives up into the back of a trailer and then uses suction to grab stacked boxes and put them onto a conveyor belt. We’ve written about a couple of the companies doing this, and there are even more out there. It’s easy to understand why—trailer unloading involves a fairly structured and controlled environment with a very repetitive task, it’s a hard job that sucks for humans, and there’s an enormous amount of demand.

While it’s likely true that there’s enough room for a whole bunch of different robotics companies in the trailer unloading space, a given customer is probably only going to pick one, and they’re going to pick the one that offers the right combination of safety, capability, and cost. Anyware Robotics thinks they have that mix, aided by a box handling solution that is both very clever and so obvious that I’m wondering why I didn’t think of it myself.

The overall design of Pixmo itself is fairly standard as far as trailer unloading robots go, but some of the details are interesting. We’re told that Pixmo is the only trailer unloading system that integrates a heavy-payload collaborative arm, actually a fairly new commercial arm from Fanuc. This means that Anyware Robotics doesn’t have to faff about with their own hardware, and also that their robot is arguably safer, being ISO certified safe to work directly with people. The base is custom, but Anyware is contracting it out to a big robotics OEM.

“We’ve put a lot of effort into making sure that most of the components of our robot are off-the-shelf,” co-founder and CEO Thomas Tang tells us. “There are already so many mature and cost-efficient suppliers that we want to offload the supply chain, the certification, the reliability testing onto someone else’s shoulders.” And while there are a selection of automated mobile robots (AMRs) out there that seem like they could get the job done, the problem is that they’re all designed for flat surfaces, and getting into and out of the back of the trailer often involves a short, steep ramp, hence the need for their own design. Even with the custom base, Tang says that Pixmo is very cost efficient, and the company predicts that it will be approximately one third the cost of other solutions with a payback of about 24 months.

But here’s the really clever bit:

Anyware Robotics Pixmo Trailer Unloading

That conveyor system in front of the boxes is an add-on that’s used in support of Pixmo. There are two benefits here: first, having the conveyor add-on aligned with the base of a box minimizes the amount of lifting that Pixmo has to do. This allows Pixmo to handle boxes of up to 65 lbs with a lift-and-slide technique, putting it at the top end of trailer unloading robot payload. And the second benefit is that the add-on system decreases the distance that Pixmo has to move the box to just about as small as it can possibly be, eliminating the need for the arm to rotate around to place a box on a conveyor next to or behind itself. Lowering this cycle time means that Pixmo can achieve a throughput of up to 1,000 boxes per hour—about one box every four seconds, which the Internet suggests is quite fast, even for a professional human. Anyware Robotics is introducing this add-on system at MODEX next week, and they have a patent pending on the idea.

This seems like such a simple, useful idea that I asked Tang why they were the first ones to come up with it. “In robotics startups, there tends to be a legacy mindset issue,” Tang told me. “When people have been working on robot arms for so many years, we just think about how to use robot arms to solve everything. That’s maybe that’s the reason why other companies didn’t come up with this solution.” Tang says that Anyware started with much more complicated add-on designs before finding this solution. “Usually it’s the most simple solution that has the most trial and error behind it.”

Anyware Robotics is focused on trailer unloading for now, but Pixmo could easily be adapted for palletizing and depalletizing or somewhat less easily for other warehouse tasks like order picking or machine tending. But why stop there? A mobile manipulator can (theoretically) do it all (almost), and that’s exactly what Tang wants:

At just over one year old, Anyware has already managed to complete a pilot program (and convert it to a purchase order). They’re currently in the middle of several other pilot programs with leading third-party logistics providers, and they expect to spend the next several months focusing on productization with the goal of releasing the first commercial version of Pixmo by July of this year.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

Figure has raised a US $675 million Series B, valuing the company at $2.6 billion.

[ Figure ]

Meanwhile, here’s how things are going at Agility Robotics, whose last raise was a $150 million Series B in April of 2022.

[ Agility Robotics ]

Also meanwhile, here’s how things are going at Sanctuary AI, whose last raise was a $58.5 million Series A in March of 2022.

[ Sanctuary AI ]

[ UBTECH ]

Henry Evans keeps working hard to make robots better, this time with the assistance of researchers from Carnegie Mellon University.

[ HAT2 ] via [ CMU ]

Watch this video for the three glorious seconds at the end.

[ Tech United ]

Oh, it’s HOOSKvarna, not HUSKvarna.

[ Husqvarna ] via [ Engadget ]

[ Engineered Arts ]

[ NYU ]

[ Boston Dynamics ]

[ Freeform Robotics ]

[ DLR ]

[ CYBATHLON ]

[ WeRobotics ]

In case you’d forgotten, Amazon has a lot of robots.

[ Amazon Robotics ]

[ ABB ]

[ Ingenuity Labs ]

Today, Figure is announcing an astonishing US $675 million Series B raise, which values the company at an even more astonishing $2.6 billion. Figure is one of the companies working towards a multi or general purpose (depending on who you ask) bipedal or humanoid (depending on who you ask) robot. The astonishing thing about this valuation is that Figure’s robot is still very much in the development phase—although they’re making rapid progress, which they demonstrate in a new video posted this week.

This round of funding comes from Microsoft, OpenAI Startup Fund, Nvidia, Jeff Bezos (through Bezos Expeditions), Parkway Venture Capital, Intel Capital, Align Ventures, and ARK Invest. Figure says that they’re going to use this new capital “for scaling up AI training, robot manufacturing, expanding engineering headcount, and advancing commercial deployment efforts.” In addition, Figure and OpenAI will be collaborating on the development of “next generation AI models for humanoid robots” which will “help accelerate Figure’s commercial timeline by enhancing the capabilities of humanoid robots to process and reason from language.”

As far as that commercial timeline goes, here’s the most recent update:

Figure

And to understand everything that’s going on here, we sent a whole bunch of questions to Jenna Reher, Senior Robotics/AI Engineer at Figure.

What does “fully autonomous” mean, exactly?

Jenna Reher: In this case, we simply put the robot on the ground and hit go on the task with no other user input. What you see is using a learned vision model for bin detection that allows us to localize the robot relative to the target bin and get the bin pose. The robot can then navigate itself to within reach of the bin, determine grasp points based on the bin pose, and detect grasp success through the measured forces on the hands. Once the robot turns and sees the conveyor the rest of the task rolls out in a similar manner. By doing things in this way we can move the bins and conveyor around in the test space or start the robot from a different position and still complete the task successfully.

How many takes did it take to get this take?

Reher: We’ve been running this use case consistently for some time now as part of our work in the lab, so we didn’t really have to change much for the filming here. We did two or three practice runs in the morning and then three filming takes. All of the takes were successful, so the extras were to make sure we got the cleanest one to show.

What’s back in the Advanced Actuator Lab?

Reher: We have an awesome team of folks working on some exciting custom actuator designs for our future robots, as well as supporting and characterizing the actuators that went into our current robots.

That’s a very specific number for “speed vs human.” Which human did you measure the robot’s speed against?

Reher: We timed Brett [Adcock, founder of Figure] and a few poor engineers doing the task and took the average to get a rough baseline. If you are observant, that seemingly over-specific number is just saying we’re at 1/6 human speed. The main point that we’re trying to make here is that we are aware we are currently below human speed, and it’s an important metric to track as we improve.

What’s the tether for?

Reher: For this task we currently process the camera data off-robot while all of the behavior planning and control happens onboard in the computer that’s in the torso. Our robots should be fully tetherless in the near future as we finish packaging all of that onboard. We’ve been developing behaviors quickly in the lab here at Figure in parallel to all of the other systems engineering and integration efforts happening, so hopefully folks notice all of these subtle parallel threads converging as we try to release regular updates.

How the heck do you keep your robotics lab so clean?

Reher: Everything we’ve filmed so far is in our large robot test lab, so it’s a lot easier to keep the area clean when people’s desks aren’t intruding in the space. Definitely no guarantees on that level of cleanliness if the camera were pointed in the other direction!

Is the robot in the background doing okay?

Reher: Yes! The other robot was patiently standing there in the background, waiting for the filming to finish up so that our manipulation team could get back to training it to do more manipulation tasks. We hope we can share some more developments with that robot as the main star in the near future.

What would happen if I put a single bowling ball into that tote?

Reher: A bowling ball is particularly menacing to this task primarily due to the moving mass, in addition to the impact if you are throwing it in. The robot would in all likelihood end up dropping the tote, stay standing, and abort the task. With what you see here, we assume that the mass of the tote is known a-priori so that our whole body controller can compensate for the external forces while tracking the manipulation task. Reacting to and estimating larger unknown disturbances such as this is a challenging problem, but we’re definitely working on it.

Tell me more about that very zen arm and hand pose that the robot adopts after putting the tote on the conveyor.

Reher: It does look kind of zen! If you re-watch our coffee video you’ll notice the same pose after the robot gets things brewing. This is a reset pose that our controller will go into between manipulation tasks while the robot is awaiting commands to execute either an engineered behavior or a learned policy.

Are the fingers less fragile than they look?

Reher: They are more robust than they look, but not impervious to damage by any means. The design is pretty modular which is great, meaning that if we damage one or two fingers there is a small number of parts to swap to get everything back up and running. The current fingers won’t necessarily survive a direct impact from a bad fall, but can pick up totes and do manipulation tasks all day without issues.

Is the Figure logo footsteps?

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

This work, by Philip Arm, Mayank Mittal, Hendrik Kolvenbach, and Marco Hutter from ETHZ RSL, will be presented at the IEEE International Conference on Robotics and Automation (ICRA 2024) in May in Japan (see events calendar above).

[ Pedipulate ]

I learned a new word today! “Stigmergy.” Stigmergy is a kind of group coordination that’s based on environmental modification. Like, when insects leave pheromone trails, they’re not directly sending messages to other individuals, but as a group the ants are able to manifest surprisingly complex coordinated behaviors. Cool, right? Researchers are IRIDIA are exploring the possibilities for robots using stigmergy with a cool ‘artificial pheromone’ system using a UV-sensitive surface.

“Automatic design of stigmergy-based behaviors for robot swarms,” by Muhammad Salman, David Garzón Ramos, and Mauro Birattari, is published in the journal Communications Engineering.

[ Nature ] via [ IRIDIA ]

Thanks, David!

[ IHMC ]

That IHMC video reminded me of the SAFFiR program for Shipboard Autonomous Firefighting Robots, which is responsible for a bunch of really cool research in partnership with the United States Naval Research Laboratory. NRL did some interesting stuff with Nexi robots from MIT and made their own videos. That effort I think didn’t get nearly enough credit for being very entertaining while communicating important robotics research.

[ NRL ]

I want more robot videos with this energy.

[ MIT CSAIL ]

[ ANYbotics ]

This week is Grain Bin Safety Week, and Grain Weevil is here to help.

[ Grain Weevil ]

Oof, this is some heavy, heavy deep-time stuff.

[ Onkalo ]

And now, this.

[ RozenZebet ]

That moment at 1:07 is so relatable.

[ Hawkeye ]

Wing would really prefer that if you find one of their drones on the ground, you don’t run off with it.

[ Wing ]

[ DFKI ]

[ Paper ]

[ MIT ]

Lex Fridman interviews Marc Raibert.

[ Lex Fridman ]

Dina Genkina: Hi. I’m Dina Genkina for IEEE Spectrum‘s Fixing the Future. Before we start, I want to tell you that you can get the latest coverage from some of Spectrum’s most important beeps, including AI, Change, and Robotics, by signing up for one of our free newsletters. Just go to spectrum.ieee.org\newsletters to subscribe. Today, a guest is Dr. Benji Maruyama, a Principal Materials Research Engineer at the Air Force Research Laboratory, or AFRL. Dr. Maruyama is a materials scientist, and his research focuses on carbon nanotubes and making research go faster. But he’s also a man with a dream, a dream of a world where science isn’t something done by a select few locked away in an ivory tower, but something most people can participate in. He hopes to start what he calls the billion scientist movement by building AI-enabled research robots that are accessible to all. Benji, thank you for coming on the show.

Benji Maruyama: Thanks, Dina. Great to be with you. I appreciate the invitation.

Genkina: Yeah. So let’s set the scene a little bit for our listeners. So you advocate for this billion scientist movement. If everything works amazingly, what would this look like? Paint us a picture of how AI will help us get there.

Maruyama: Right, great. Thanks. Yeah. So one of the things as you set the scene there is right now, to be a scientist, most people need to have access to a big lab with very expensive equipment. So I think top universities, government labs, industry folks, lots of equipment. It’s like a million dollars, right, to get one of them. And frankly, just not that many of us have access to those kinds of instruments. But at the same time, there’s probably a lot of us who want to do science, right? And so how do we make it so that anyone who wants to do science can try, can have access to instruments so that they can contribute to it. So that’s the basics behind citizen science or democratization of science so that everyone can do it. And one way to think of it is what happened with 3D printing. It used to be that in order to make something, you had to have access to a machine shop or maybe get fancy tools and dyes that could cost tens of thousands of dollars a pop. Or if you wanted to do electronics, you had to have access to very expensive equipment or services. But when 3D printers came along and became very inexpensive, all of a sudden now, anyone with access to a 3D printer, so maybe in a school or a library or a makerspace could print something out. And it could be something fun, like a game piece, but it could also be something that got you to an invention, something that was maybe useful to the community, was either a prototype or an actual working device.

And so really, 3D printing democratized manufacturing, right? It made it so that many more of us could do things that before only a select few could. And so that’s where we’re trying to go with science now, is that instead of only those of us who have access to big labs, we’re building research robots. And when I say we, we’re doing it, but now there are a lot of others who are doing it as well, and I’ll get into that. But the example that we have is that we took a 3D printer that you can buy off the internet for less than $300. Plus a couple of extra parts, a webcam, a Raspberry Pi board, and a tripod really, so only four components. You can get them all for $300. Load them with open-source software that was developed by AFIT, the Air Force Institute of Technology. So Burt Peterson and Greg Captain [inaudible]. We worked together to build this fully autonomous 3D printing robot that taught itself how to print to better than manufacturer’s specifications. So that was a really fun advance for us, and now we’re trying to take that same idea and broaden it. So I’ll turn it back over to you.

Genkina: Yeah, okay. So maybe let’s talk a little bit about this automated research robot that you’ve made. So right now, it works with a 3D printer, but is the big picture that one day it’s going to give people access to that million dollar lab? How would that look like?

Maruyama: Right, so there are different models out there. One, we just did a workshop at the University of— sorry, North Carolina State University about that very problem, right? So there’s two models. One is to get low-cost scientific tools like the 3D printer. There’s a couple of different chemistry robots, one out of University of Maryland and NIST, one out of University of Washington that are in the sort of 300 to 1,000 dollars range that makes it accessible. The other part is kind of the user facility model. So in the US, the Department of Energy National Labs have many user facilities where you can apply to get time on very expensive instruments. Now we’re talking tens of millions. For example, Brookhaven has a synchrotron light source where you can sign up and it doesn’t cost you any money to use the facility. And you can get days on that facility. And so that’s already there, but now the advances are that by using this, autonomy, autonomous closed loop experimentation, that the work that you do will be much faster and much more productive. So, for example, on ARES, our Autonomous Research System at AFRL, we actually were able to do experiments so fast that a professor who came into my lab said, it just took me aside and said, “Hey, Benji, in a week’s worth of time, I did a dissertation’s worth of research.” So maybe five years worth of research in a week. So imagine if you keep doing that week after week after week, how fast research goes. So it’s very exciting.

Genkina: Yeah, so tell us a little bit about how that works. So what’s this system that has sped up five years of research into a week and made graduate students obsolete? Not yet, not yet. How does that work? Is that the 3D printer system or is that a—

Maruyama: So we started with our system to grow carbon nanotubes. And I’ll say, actually, when we first thought about it, your comment about graduate students being absolute— obsolete, sorry, is interesting and important because, when we first built our system that worked it 100 times faster than normal, I thought that might be the case. We called it sort of graduate student out of the loop. But when I started talking with people who specialize in autonomy, it’s actually the opposite, right? It’s actually empowering graduate students to go faster and also to do the work that they want to do, right? And so just to digress a little bit, if you think about farmers before the Industrial Revolution, what were they doing? They were plowing fields with oxen and beasts of burden and hand plows. And it was hard work. And now, of course, you wouldn’t ask a farmer today to give up their tractor or their combine harvester, right? They would say, of course not. So very soon, we expect it to be the same for researchers, that if you asked a graduate student to give up their autonomous research robot five years from now, they’ll say, “Are you crazy? This is how I get my work done.”

But for our original ARES system, it worked on the synthesis of carbon nanotubes. So that meant that what we’re doing is trying to take this system that’s been pretty well studied, but we haven’t figured out how to make it at scale. So at hundreds of millions of tons per year, sort of like polyethylene production. And part of that is because it’s slow, right? One experiment takes a day, but also because there are just so many different ways to do a reaction, so many different combinations of temperature and pressure and a dozen different gases and half the periodic table as far as the catalyst. It’s just too much to just brute force your way through. So even though we went from experiments where we could do 100 experiments a day instead of one experiment a day, just that combinatorial space was vastly overwhelmed our ability to do it, even with many research robots or many graduate students. So the idea of having artificial intelligence algorithms that drive the research is key. And so that ability to do an experiment, see what happened, and then analyze it, iterate, and constantly be able to choose the optimal next best experiment to do is where ARES really shines. And so that’s what we did. ARES taught itself how to grow carbon nanotubes at controlled rates. And we were the first ones to do that for material science in our 2016 publication.

Genkina: That’s very exciting. So maybe we can peer under the hood a little bit of this AI model. How does the magic work? How does it pick the next best point to take and why it’s better than you could do as a graduate student or researcher?

Maruyama: Yeah, and so I think it’s interesting, right? In science, a lot of times we’re taught to hold everything constant, change one variable at a time, search over that entire space, see what happened, and then go back and try something else, right? So we reduce it to one variable at a time. It’s a reductionist approach. And that’s worked really well, but a lot of the problems that we want to go after are simply too complex for that reductionist approach. And so the benefit of being able to use artificial intelligence is that high dimensionality is no problem, right? Tens of dimensions search over very complex high-dimensional parameter space, which is overwhelming to humans, right? Is just basically bread and butter for AI. The other part to it is the iterative part. The beauty of doing autonomous experimentation is that you’re constantly iterating. You’re constantly learning over what just happened. You might also say, well, not only do I know what happened experimentally, but I have other sources of prior knowledge, right? So for example, ideal gas law says that this should happen, right? Or Gibbs phase rule might say, this can happen or this can’t happen. So you can use that prior knowledge to say, “Okay, I’m not going to do those experiments because that’s not going to work. I’m going to try here because this has the best chance of working.”

And within that, there are many different machine learning or artificial intelligence algorithms. Bayesian optimization is a popular one to help you choose what experiment is best. There’s also new AI that people are trying to develop to get better search.

Genkina: Cool. And so the software part of this autonomous robot is available for anyone to download, which is also really exciting. So what would someone need to do to be able to use that? Do they need to get a 3D printer and a Raspberry Pi and set it up? And what would they be able to do with it? Can they just build carbon nanotubes or can they do more stuff?

Maruyama: Right. So what we did, we built ARES OS, which is our open source software, and we’ll make sure to get you the GitHub link so that anyone can download it. And the idea behind ARES OS is that it provides a software framework for anyone to build their own autonomous research robot. And so the 3D printing example will be out there soon. But it’s the starting point. Of course, if you want to build your own new kind of robot, you still have to do the software development, for example, to link the ARES framework, the core, if you will, to your particular hardware, maybe your particular camera or 3D printer, or pipetting robot, or spectrometer, whatever that is. We have examples out there and we’re hoping to get to a point where it becomes much more user-friendly. So having direct Python connects so that you don’t— currently it’s programmed in C#. But to make it more accessible, we’d like it to be set up so that if you can do Python, you can probably have good success in building your own research robot.

Genkina: Cool. And you’re also working on a educational version of this, I understand. So what’s the status of that and what’s different about that version?

Maruyama: Yeah, right. So the educational version is going to be-- its sort of composition of a combination of hardware and software. So what we’re starting with is a low-cost 3D printer. And we’re collaborating now with the University at Buffalo, Materials Design Innovation Department. And we’re hoping to build up a robot based on a 3D printer. And we’ll see how it goes. It’s still evolving. But for example, it could be based on this very inexpensive $200 3D printer. It’s an Ender 3D printer. There’s another printer out there that’s based on University of Washington’s Jubilee printer. And that’s a very exciting development as well. So professors Lilo Pozzo and Nadya Peek at the University of Washington built this Jubilee robot with that idea of accessibility in mind. And so combining our ARES OS software with their Jubilee robot hardware is something that I’m very excited about and hope to be able to move forward on.

Genkina: What’s this Jubilee 3D printer? How is it different from a regular 3D printer?

Maruyama: It’s very open source. Not all 3D printers are open source and it’s based on a gantry system with interchangeable heads. So for example, you can get not just a 3D printing head, but other heads that might do things like do indentation, see how stiff something is, or maybe put a camera on there that can move around. And so it’s the flexibility of being able to pick different heads dynamically that I think makes it super useful. For the software, right, we have to have a good, accessible, user-friendly graphical user interface, a GUI. That takes time and effort, so we want to work on that. But again, that’s just the hardware software. Really to make ARES a good educational platform, we need to make it so that a teacher who’s interested can have the lowest activation barrier possible, right? We want she or he to be able to pull a lesson plan off of the internet, have supporting YouTube videos, and actually have the material that is a fully developed curriculum that’s mapped against state standards.

So that, right now, if you’re a teacher who— let’s face it, teachers are already overwhelmed with all that they have to do, putting something like this into their curriculum can be a lot of work, especially if you have to think about, well, I’m going to take all this time, but I also have to meet all of my teaching standards, all the state curriculum standards. And so if we build that out so that it’s a matter of just looking at the curriculum and just checking off the boxes of what state standards it maps to, then that makes it that much easier for the teacher to teach.

Genkina: Great. And what do you think is the timeline? Do you expect to be able to do this sometime in the coming year?

Maruyama: That’s right. These things always take longer than hoped for than expected, but we’re hoping to do it within this calendar year and very excited to get it going. And I would say for your listeners, if you’re interested in working together, please let me know. We’re very excited about trying to involve as many people as we can.

Genkina: Great. Okay, so you have the educational version, and you have the more research geared version, and you’re working on making this educational version more accessible. Is there something with the research version that you’re working on next, how you’re hoping to upgrade it, or is there something you’re using it for right now that you’re excited about?

There’s a number of things that we are very excited about the possibility of carbon nanotubes being produced at very large scale. So right now, people may remember carbon nanotubes as that great material that sort of never made it and was very overhyped. But there’s a core group of us who are still working on it because of the important promise of that material. So it’s material that is super strong, stiff, lightweight, electrically conductive. Much better than silicon as a digital electronics compute material. All of those great things, except we’re not making it at large enough scale. It’s actually used pretty significantly in lithium-ion batteries. It’s an important application. But other than that, it’s sort of like where’s my flying car? It’s never panned out. But there’s, as I said, a group of us who are working to really produce carbon nanotubes at much larger scale. So large scale for nanotubes now is sort of in the kilogram or ton scale. But what we need to get to is hundreds of millions of tons per year production rates. And why is that? Well, there’s a great effort that came out of ARPA-E. So the Department of Energy Advanced Research Projects Agency and the E is for Energy in that case.

So they funded a collaboration between Shell Oil and Rice University to pyrolyze methane, so natural gas into hydrogen for the hydrogen economy. So now that’s a clean burning fuel plus carbon. And instead of burning the carbon to CO2, which is what we now do, right? We just take natural gas and feed it through a turbine and generate electric power instead of— and that, by the way, generates so much CO2 that it’s causing global climate change. So if we can do that pyrolysis at scale, at hundreds of millions of tons per year, it’s literally a save the world proposition, meaning that we can avoid so much CO2 emissions that we can reduce global CO2 emissions by 20 to 40 percent. And that is the save the world proposition. It’s a huge undertaking, right? That’s a big problem to tackle, starting with the science. We still don’t have the science to efficiently and effectively make carbon nanotubes at that scale. And then, of course, we have to take the material and turn it into useful products. So the batteries is the first example, but thinking about replacing copper for electrical wire, replacing steel for structural materials, aluminum, all those kinds of applications. But we can’t do it. We can’t even get to that kind of development because we haven’t been able to make the carbon nanotubes at sufficient scale.

So I would say that’s something that I’m working on now that I’m very excited about and trying to get there, but it’s going to take some good developments in our research robots and some very smart people to get us there.

Genkina: Yeah, it seems so counterintuitive that making everything out of carbon is good for lowering carbon emissions, but I guess that’s the break.

Maruyama: Yeah, it is interesting, right? So people talk about carbon emissions, but really, the molecule that’s causing global warming is carbon dioxide, CO2, which you get from burning carbon. And so if you take that methane and parallelize it to carbon nanotubes, that carbon is now sequestered, right? It’s not going off as CO2. It’s staying in solid state. And not only is it just not going up into the atmosphere, but now we’re using it to replace steel, for example, which, by the way, steel, aluminum, copper production, all of those things emit lots of CO2 in their production, right? They’re energy intensive as a material production. So it’s kind of ironic.

Genkina: Okay, and are there any other research robots that you’re excited about that you think are also contributing to this democratization of science process?

Maruyama: Yeah, so we talked about Jubilee, the NIST robot, which is from Professor Ichiro Takeuchi at Maryland and Gilad Kusne at NIST, National Institute of Standards and Technology. Theirs is fun too. It’s LEGO as. So it’s actually based on a LEGO robotics platform. So it’s an actual chemistry robot built out of Legos. So I think that’s fun as well. And you can imagine, just like we have LEGO robot competitions, we can have autonomous research robot competitions where we try and do research through these robots or competitions where everybody sort of starts with the same robot, just like with LEGO robotics. So that’s fun as well. But I would say there’s a growing number of people doing these kinds of, first of all, low-cost science, accessible science, but in particular low-cost autonomous experimentation.

Genkina: So how far are we from a world where a high school student has an idea and they can just go and carry it out on some autonomous research system at some high-end lab?

Maruyama: That’s a really good question. I hope that it’s going to be in 5 to 10 years, that it becomes reasonably commonplace. But it’s going to take still some significant investment to get this going. And so we’ll see how that goes. But I don’t think there are any scientific impediments to getting this done. There is a significant amount of engineering to be done. And sometimes we hear, oh, it’s just engineering. The engineering is a significant problem. And it’s work to get some of these things accessible, low cost. But there are lots of great efforts. There are people who have used CDs, compact discs to make spectrometers out of. There are lots of good examples of citizen science out there. But it’s, I think, at this point, going to take investment in software, in hardware to make it accessible, and then importantly, getting students really up to speed on what AI is and how it works and how it can help them. And so I think it’s actually really important. So again, that’s the democratization of science is if we can make it available to everyone and accessible, then that helps people, everyone contribute to science. And I do believe that there are important contributions to be made by ordinary citizens, by people who aren’t you know PhDs working in a lab.

And I think there’s a lot of science out there to be done. If you ask working scientists, almost no one has run out of ideas or things they want to work on. There’s many more scientific problems to work on than we have the time where people are funding to work on. And so if we make science cheaper to do, then all of a sudden, more people can do science. And so those questions start to be resolved. And so I think that’s super important. And now we have, instead of, just those of us who work in big labs, you have millions, tens of millions, up to a billion people, that’s the billion scientist idea, who are contributing to the scientific community. And that, to me, is so powerful that many more of us can contribute than just the few of us who do it right now.

Genkina: Okay, that’s a great place to end on, I think. So, today we spoke to Dr. Benji Maruyama, a material scientist at AFRL, about his efforts to democratize scientific discovery through automated research robots. For IEEE Spectrum, I’m Dina Genkina, and I hope you’ll join us next time on Fixing the Future.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

Just like a real human, Acrobot will sometimes kick you in the face.

[ Acrobotics ]

Thanks, Elizabeth!

You had me at “wormlike, limbless robots.”

[ GitHub ] via [ Georgia Tech ]

I would say that in the middle of a rainstorm is probably the best time to start a fire that you expect to be extinguished by a robot.

[ IHMC ]

[ Boston Dynamics ]

This is pretty cool: BruBotics is testing its self-healing robotics gripper technology on commercial grippers from Festo.

[ Paper ] via [ BruBotics ]

Thanks, Bram!

You should read our in-depth article on Stretch 3, so if you haven’t yet, consider this as just a teaser.

[ Hello Robot ]

[ Project ]

Thanks, Jihong!

Tony Punnoose Valayil from the Bulgarian Academy of Sciences Institute of Robotics wrote in to share some very low-cost hand-rehabilitation robots for home use.

[ BAS ]

Some very impressive work here from the Norwegian University of Science and Technology (NTNU), showing a drone tracking its position using radar and lidar-based odometry in some nightmare (for robots) environments, including a long tunnel that looks the same everywhere and a hallway full of smoke.

[ Paper ] via [ GitHub ]

I’m sorry, but people should really know better than to make videos like this for social robot crowdfunding by now.

It’s on Kickstarter for about $300, and the fact that it’s been funded so quickly tells me that people have already forgotten about the social robotpocalypse.

[ Kickstarter ]

[ Boston Dynamics ]

A lot has happened in robotics over the last year. Everyone is wondering how AI will transform robotics, and everyone else is wondering whether humanoids are going to blow it or not, and the rest of us are busy trying not to get completely run over as things shake out however they’re going to shake out.

Meanwhile, over at Hello Robot, they’ve been focused on making their Stretch robot do useful things while also being affordable and reliable and affordable and expandable and affordable and community-friendly and affordable. Which are some really hard and important problems that can sometimes get overwhelmed by flashier things.

Today, Hello Robot is announcing Stretch 3, which provides a suite of upgrades to what they (quite accurately) call “the world’s only lightweight, capable, developer-friendly mobile manipulator.” And impressively, they’ve managed to do it without forgetting about that whole “affordable” part.

Hello Robot

Stretch 3 looks about the same as the previous versions, but there are important upgrades that are worth highlighting. The most impactful: Stretch 3 now comes with the dexterous wrist kit that used to be an add-on, and it now also includes an Intel Realsense D405 camera mounted right behind the gripper, which is a huge help for both autonomy and remote teleoperation—a useful new feature shipping with Stretch 3 that’s based on research out of Maya Cakmak’s lab at the University of Washington, in Seattle. This is an example of turning innovation from the community of Stretch users into product features, a product-development approach that seems to be working well for Hello Robot.

“We’ve really been learning from our community,” says Hello Robot cofounder and CEO Aaron Edsinger. “In the past year, we’ve seen a real uptick in publications, and it feels like we’re getting to this critical-mass moment with Stretch. So with Stretch 3, it’s about implementing features that our community has been asking us for.”

“When we launched, we didn’t have a dexterous wrist at the end as standard, because we were trying to start with truly the minimum viable product,” says Hello Robot cofounder and CTO Charlie Kemp. “And what we found is that almost every order was adding the dexterous wrist, and by actually having it come in standard, we’ve been able to devote more attention to it and make it a much more robust and capable system.”

Kemp says that having Stretch do everything right out of the box (with Hello Robot support) makes a big difference for their research customers. “Making it easier for people to try things—we’ve learned to really value that, because the more steps that people have to go through to experience it, the less likely they are to build on it.” In a research context, this is important because what you’re really talking about is time: The more time people spend just trying to make the robot function, the less time they’ll spend getting the robot to do useful things.

Hello Robot

At this point, you may be thinking of Stretch as a research platform. Or you may be thinking of Stretch as a robot for people with disabilities, if you read our November 2023 cover story about Stretch and Henry and Jane Evans. And the robot is definitely both of those things. But Hello Robot stresses that these specific markets are not their end goal—they see Stretch as a generalist mobile manipulator with a future in the home, as suggested by this Stretch 3 promo video:

Hello Robot

Dishes, laundry, bubble cannons: All of these are critical to the functionality of any normal household. “Stretch is an inclusive robot,” says Kemp. “It’s not just for older adults or people with disabilities. We want a robot that can be beneficial for everyone. Our vision, and what we believe will really happen, whether it’s us or someone else, is that there is going to be a versatile, general-purpose home robot. Right now, clearly, our market is not yet consumers in the home. But that’s where we want to go.”

Robots in the home have been promised for decades, and with the notable exception of the Roomba, there has not been a lot of success. The idea of a robot that could handle dishes or laundry is tempting, but is it near-term or medium-term realistic? Edsinger, who has been at this whole robots thing for a very long time, is an optimist about this, and about the role that Stretch will play. “There are so many places where you can see the progress happening—in sensing, in manipulation,” Edsinger says. “I can imagine those things coming together now in a way that I could not have 5 to 10 years ago, when it seemed so incredibly hard.”

“We’re very pragmatic about what is possible. And I think that we do believe that things are changing faster than we anticipated—10 years ago, I had a pretty clear linear path in mind for robotics, but it’s hard to really imagine where we’ll be in terms of robot capabilities 10 years from now.” —Aaron Edsinger, Hello Robot

I’d say that it’s still incredibly hard, but Edsinger is right that a lot of the pieces do seem to be coming together. Arguably, the hardware is the biggest challenge here, because working in a home puts heavy constraints on what kind of hardware you’re able to use. You’re not likely to see a humanoid in a home anytime soon, because they’d actually be dangerous, and even a quadruped is likely to be more trouble than it’s worth in a home environment. Hello Robot is conscious of this, and that’s been one of the main drivers of the design of Stretch.

“I think the portability of Stretch is really worth highlighting because there’s just so much value in that which is maybe not obvious,” Edsinger tells us. Being able to just pick up and move a mobile manipulator is not normal. Stretch’s weight (24.5 kilograms) is almost trivial to work with, in sharp contrast with virtually every other mobile robot with an arm: Stretch fits into places that humans fit into, and manages to have a similar workspace as well, and its bottom-heavy design makes it safe for humans to be around. It can’t climb stairs, but it can be carried upstairs, which is a bigger deal than it may seem. It’ll fit in the back of a car, too. Stretch is built to explore the world—not just some facsimile of the world in a research lab.

“NYU students have been taking Stretch into tens of homes around New York,” says Edsinger. “They carried one up a four-story walk-up. This enables real in-home data collection. And this is where home robots will start to happen—when you can have hundreds of these out there in homes collecting data for machine learning.”

“That’s where the opportunity is,” adds Kemp. “It’s that engagement with the world about where to apply the technology beneficially. And if you’re in a lab, you’re not going to find it.”

We’ve seen some compelling examples of this recently, with Mobile ALOHA. These are robots learning to be autonomous by having humans teleoperate them through common household skills. But the system isn’t particularly portable, and it costs nearly US $32,000 in parts alone. Don’t get me wrong: I love the research. It’s just going to be difficult to scale, and in order to collect enough data to effectively tackle the world, scale is critical. Stretch is much easier to scale, because you can just straight up buy one.

Or two! You may have noticed that some of the Stretch 3 videos have two robots in them, collaborating with each other. This is not yet autonomous, but with two robots, a single human (or a pair of humans) can teleoperate them as if they were effectively a single two-armed robot:

Hello Robot

Essentially, what you’ve got here is a two-armed robot that (very intentionally) has nothing to do with humanoids. As Kemp explains: “We’re trying to help our community and the world see that there is a different path from the human model. We humans tend to think of the preexisting solution: People have two arms, so we think, well, I’m going to need to have two arms on my robot or it’s going to have all these issues.” Kemp points out that robots like Stretch have shown that really quite a lot of things can be done with only one arm, but two arms can still be helpful for a substantial subset of common tasks. “The challenge for us, which I had just never been able to find a solution for, was how you get two arms into a portable, compact, affordable lightweight mobile manipulator. You can’t!”

But with two Stretches, you have not only two arms but also two shoulders that you can put wherever you want. Washing a dish? You’ll probably want two arms close together for collaborative manipulation. Making a bed? Put the two arms far apart to handle both sides of a sheet at once. It’s a sort of distributed on-demand bimanual manipulation, which certainly adds a little bit of complexity but also solves a bunch of problems when it comes to practical in-home manipulation. Oh—and if those teleop tools look like modified kitchen tongs, that’s because they’re modified kitchen tongs.

Of course, buying two Stretch robots is twice as expensive as buying a single Stretch robot, and even though Stretch 3’s cost of just under $25,000 is very inexpensive for a mobile manipulator and very affordable in a research or education context, we’re still pretty far from something that most people would be able to afford for themselves. Hello Robot says that producing robots at scale is the answer here, which I’m sure is true, but it can be a difficult thing for a small company to achieve.

Moving slowly toward scale is at least partly intentional, Kemp tells us. “We’re still in the process of discovering Stretch’s true form—what the robot really should be. If we tried to scale to make lots and lots of robots at a much lower cost before we fundamentally understood what the needs and challenges were going to be, I think it would be a mistake. And there are many gravestones out there for various home-robotics companies, some of which I truly loved. We don’t want to become one of those.”

This is not to say that Hello Robot isn’t actively trying to make Stretch more affordable, and Edsinger suggests that the next iteration of the robot will be more focused on that. But—and this is super important—Kemp tells us that Stretch has been, is, and will continue to be sustainable for Hello Robot: “We actually charge what we should be charging to be able to have a sustainable business.” In other words, Hello Robot is not relying on some nebulous scale-defined future to transition into a business model that can develop, sell, and support robots. They can do that right now while keeping the lights on. “Our sales have enough margin to make our business work,” says Kemp. “That’s part of our discipline.”

Odorigui is a type of Japanese cuisine in which people consume live seafood while it’s still moving, making movement part of the experience. You may have some feelings about this (I definitely do), but from a research perspective, getting into what those feelings are and what they mean isn’t really practical. To do so in a controlled way would be both morally and technically complicated, which is why Japanese researchers have started developing robots that can be eaten as they move, wriggling around in your mouth as you chomp down on them. Welcome to HERI: Human-Edible Robot Interaction.

That happy little robot that got its head ripped off by a hungry human (who, we have to say, was exceptionally polite about it) is made primarily of gelatin, along with sugar and apple juice for taste. After all the ingredients were mixed, it was poured into a mold and refrigerated for 12 hours to set, with the resulting texture ending up like a chewy gummy candy. The mold incorporated a couple of air chambers into the structure of the robot, which were hooked up to pneumatics that got the robot to wiggle back and forth.

Sixteen students at Osaka University got the chance to eat one of these wiggly little robots. The process was to put your mouth around the robot, let the robot move around in there for 10 seconds for the full experience, and then bite it off, chew, and swallow. Japanese people were chosen partly because this research was done in Japan, but also because, according to the paper, “of the cultural influences on the use of onomatopoeic terms.” In Japanese, there are terms that are useful in communicating specific kinds of textures that can’t easily be quantified.

The participants were asked a series of questions about their experience, including some heavy ones:

• Did you think what you just ate had animateness?
• Did you feel an emotion in what you just ate?
• Did you think what you just ate had intelligence?
• Did you feel guilty about what you just ate?

Oof.

Compared to a control group of students who ate the robot when it was not moving, the students who ate the moving robot were more likely to interpret it as having a “munya-munya” or “mumbly” texture, showing that movement can influence the eating experience. Analysis of question responses showed that the moving robot also caused people to perceive it as emotive and intelligent, and caused more feelings of guilt when it was consumed. The paper summarizes it pretty well: “In the stationary condition, participants perceived the robot as ‘food,’ whereas in the movement condition, they perceived it as a ‘creature.’”

The good news here is that since these robots are more like living things than non-robots, they could potentially stand in for eating live critters in a research context, say the researchers: “The utilization of edible robots in this study enabled us to examine the effects of subtle movement variations in human eating behavior under controlled conditions, a task that would be challenging to accomplish with real organisms.” There’s still more work to do to make the robots more like specific living things, but that’s the plan going forward:

Exploring the eating experience of a pneumatically-driven edible robot: Perception, taste, and texture, by Yoshihiro NakataI, Midori Ban, Ren Yamaki, Kazuya Horibe, Hideyuki Takahashi, and Hiroshi Ishiguro from The University of Electro-Communications and Osaka University, is published in PLOS One.

Just last month, Oslo, Norway-based 1X (formerly Halodi Robotics) announced a massive $100 million Series B, and clearly they’ve been putting the work in. A new video posted last week shows a [insert collective noun for humanoid robots here] of EVE android-ish mobile manipulators doing a wide variety of tasks leveraging end-to-end neural networks (pixels to actions). And best of all, the video seems to be more or less an honest one: a single take, at (appropriately) 1X speed, and full autonomy. But we still had questions! And 1X has answers.

If, like me, you had some very important questions after watching this video, including whether that plant is actually dead and the fate of the weighted companion cube, you’ll want to read this Q&A with Eric Jang, Vice President of Artificial Intelligence at 1X.

IEEE Spectrum: How many takes did it take to get this take?

Eric Jang: About 10 takes that lasted more than a minute; this was our first time doing a video like this, so it was more about learning how to coordinate the film crew and set up the shoot to look impressive.

Did you train your robots specifically on floppy things and transparent things?

Jang: Nope! We train our neural network to pick up all kinds of objects—both rigid and deformable and transparent things. Because we train manipulation end-to-end from pixels, picking up deformables and transparent objects is much easier than a classical grasping pipeline, where you have to figure out the exact geometry of what you are trying to grasp.

What keeps your robots from doing these tasks faster?

Jang: Our robots learn from demonstrations, so they go at exactly the same speed the human teleoperators demonstrate the task at. If we gathered demonstrations where we move faster, so would the robots.

How many weighted companion cubes were harmed in the making of this video?

Jang: At 1X, weighted companion cubes do not have rights.

That’s a very cool method for charging, but it seems a lot more complicated than some kind of drive-on interface directly with the base. Why use manipulation instead?

Jang: You’re right that this isn’t the simplest way to charge the robot, but if we are going to succeed at our mission to build generally capable and reliable robots that can manipulate all kinds of objects, our neural nets have to be able to do this task at the very least. Plus, it reduces costs quite a bit and simplifies the system!

What animal is that blue plush supposed to be?

Jang: It’s an obese shark, I think.

How many different robots are in this video?

Jang: 17? And more that are stationary.

How do you tell the robots apart?

Jang: They have little numbers printed on the base.

Is that plant dead?

Jang: Yes, we put it there because no CGI / 3D rendered video would ever go through the trouble of adding a dead plant.

What sort of existential crisis is the robot at the window having?

Jang: It was supposed to be opening and closing the window repeatedly (good for testing statistical significance).

If one of the robots was actually a human in a helmet and a suit holding grippers and standing on a mobile base, would I be able to tell?

Jang: I was super flattered by this comment on the Youtube video:

But if you look at the area where the upper arm tapers at the shoulder, it’s too thin for a human to fit inside while still having such broad shoulders:

Why are your robots so happy all the time? Are you planning to do more complex HRI stuff with their faces?

Jang: Yes, more complex HRI stuff is in the pipeline!

Are your robots able to autonomously collaborate with each other?

Jang: Stay tuned!

Is the skew tetromino the most difficult tetromino for robotic manipulation?

Jang: Good catch! Yes, the green one is the worst of them all because there are many valid ways to pinch it with the gripper and lift it up. In robotic learning, if there are multiple ways to pick something up, it can actually confuse the machine learning model. Kind of like asking a car to turn left and right at the same time to avoid a tree.

Everyone else’s robots are making coffee. Can your robots make coffee?

Jang: Yep! We were planning to throw in some coffee making on this video as an easter egg, but the coffee machine broke right before the film shoot and it turns out it’s impossible to get a Keurig K-Slim in Norway via next day shipping.

1X is currently hiring both AI researchers (imitation learning, reinforcement learning, large-scale training, etc) and android operators (!) which actually sounds like a super fun and interesting job. More here.

This is a guest post. The views expressed here are solely those of the author and do not represent positions of IEEE Spectrum or the IEEE.

If Disney’s history of storytelling has taught us anything, it’s to never underestimate the power of a great sidekick. Even though sidekicks aren’t the stars of the show, they provide life and energy and move the story along in important ways. It’s hard to imagine Aladdin without the Genie, or Peter Pan without Tinker Bell.

In robotics, however, solo acts proliferate. Even when multiple robots are used, they usually act in parallel. One key reason for this is that most robots are designed in ways that make direct collaboration with other robots difficult. Stiff, strong robots are more repeatable and easier to control, but those designs have very little forgiveness for the imperfections and mismatches that are inherent in coming into contact with another robot.

Having robots work together–especially if they have complementary skill sets–can open up some exciting opportunities, especially in the entertainment robotics space. At Walt Disney Imagineering, our research and development teams have been working on this idea of collaboration between robots, and we were able to show off the result of one such collaboration in Shanghai this week, when a little furry character interrupted the opening moments for the first-ever Zootopia land.

Our newest robotic character, Duke Weaselton, rolled onstage at the Shanghai Disney Resort for the first time last December, pushing a purple kiosk and blasting pop music. As seen in the video below, the audience got a kick out of watching him hop up on top of the kiosk and try to negotiate with the Chairman of Disney Experiences, Josh D’Amaro, for a new job. And of course, some new perks. After a few moments of wheeling and dealing, Duke gets gently escorted offstage by team members Richard Landon and Louis Lambie.

What might not be obvious at first is that the moment you just saw was enabled not by one robot, but by two. Duke Weaselton is the star of the show, but his dynamic motion wouldn’t be possible without the kiosk, which is its own independent, actuated robot. While these two robots are very different, by working together as one system, they’re able to do things that neither could do alone.

The character and the kiosk bring two very different kinds of motion together, and create something more than the sum of their parts in the process. The character is an expressive, bipedal robot with an exaggerated, animated motion style. It looks fantastic, but it’s not optimized for robust, reliable locomotion. The kiosk, meanwhile, is a simple wheeled system that behaves in a highly predictable way. While that’s great for reliability, it means that by itself it’s not likely to surprise you. But when we combine these two robots, we get the best of both worlds. The character robot can bring a zany, unrestrained energy and excitement as it bounces up, over, and alongside the kiosk, while the kiosk itself ensures that both robots reliably get to wherever they are going.

Harout Jarchafjian, Sophie Bowe, Tony Dohi, Bill West, Marcela de los Rios, Bob Michel, and Morgan Pope.Morgan Pope

The collaboration between the two robots is enabled by designing them to be robust and flexible, and with motions that can tolerate a large amount of uncertainty while still delivering a compelling show. This is a direct result from lessons learned from an earlier robot, one that tumbled across the stage at SXSW earlier this year. Our basic insight is that a small, lightweight robot can be surprisingly tough, and that this toughness enables new levels of creative freedom in the design and execution of a show.

This level of robustness also makes collaboration between robots easier. Because the character robot is tough and because there is some flexibility built into its motors and joints, small errors in placement and pose don’t create big problems like they might for a more conventional robot. The character can lean on the motorized kiosk to create the illusion that it is pushing it across the stage. The kiosk then uses a winch to hoist the character onto a platform, where electromagnets help stabilize its feet. Essentially, the kiosk is compensating for the fact that Duke himself can’t climb, and might be a little wobbly without having his feet secured. The overall result is a free-ranging bipedal robot that moves in a way that feels natural and engaging, but that doesn’t require especially complicated controls or highly precise mechanical design. Here’s a behind-the-scenes look at our development of these systems:

Disney Imagineering

To program Duke’s motions, our team uses an animation pipeline that was originally developed for the SXSW demo, where a designer can pose the robot by hand to create new motions. We have since developed an interface which can also take motions from conventional animation software tools. Motions can then be adjusted to adapt to the real physical constraints of the robots, and that information can be sent back to the animation tool. As animations are developed, it’s critical to retain a tight synchronization between the kiosk and the character. The system is designed so that the motion of both robots is always coordinated, while simultaneously supporting the ability to flexibly animate individual robots–or individual parts of the robot, like the mouth and eyes.

Over the past nine months, we explored a few different kinds of collaborative locomotion approaches. The GIFs below show some early attempts at riding a tricycle, skateboarding, and pushing a crate. In each case, the idea is for a robotic character to eventually collaborate with another robotic system that helps bring that character’s motions to life in a stable and repeatable way.

Disney hopes that their Judy Hopps robot will soon be able to use the help of a robotic tricycle, crate, or skateboard to enable new forms of locomotion.Morgan Pope

This demo with Duke Weaselton and his kiosk is just the beginning, says Principal R&D Imagineer Tony Dohi, who leads the project for us. “Ultimately, what we showed today is an important step towards a bigger vision. This project is laying the groundwork for robots that can interact with each other in surprising and emotionally satisfying ways. Today it’s a character and a kiosk, but moving forward we want to have multiple characters that can engage with each other and with our guests.”

Walt Disney Imagineering R&D is exploring a multi-pronged development strategy for our robotic characters. Engaging character demonstrations like Duke Weasleton focus on quickly prototyping complete experiences using immediately accessible techniques. In parallel, our research group is developing new technologies and capabilities that become the building blocks for both elevating existing experiences, and designing and delivering completely new shows. The robotics team led by Moritz Bächer shared one such building block–embodied in a highly expressive and stylized robotic walking character–at IROS in October. The capabilities demonstrated there can eventually be used to help robots like Duke Weaselton perform more flexibly, more reliably, and more spectacularly.

“Authentic character demonstrations are useful because they help inform what tools are the most valuable for us to develop,” explains Bächer. “In the end our goal is to create tools that enable our teams to produce and deliver these shows rapidly and efficiently.” This ties back to the fundamental technical idea behind the Duke Weaselton show moment–collaboration is key!

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

[ Paper ] via [ KIMLAB ]

Making money with robots has always been a struggle, but I think ALOHA 2 has figured it out.

Seriously, though, that is some impressive manipulation capability. I don’t know what that freakish panda thing is, but getting a contact lens from the package onto its bizarre eyeball was some wild dexterity.

[ ALOHA 2 ]

The least dramatic but possibly most important bit of that video is when Nadia uses her arms to help her balance against a wall, which is one of those things that humans do all the time without thinking about it. And we always appreciate being shown things that don’t go perfectly alongside things that do. The bit at the end there was Nadia not quite managing to do lateral arm raises. I can relate; that’s my reaction when I lift weights, too.

[ IHMC ]

Thanks, Robert!

The recent progress in commercial humanoids is just exhausting.

[ Unitree ]

[ Science Robotics ]

[ Ascento ]

Notable that Boston Dynamics is now saying that Atlas “gets ready for real work.” Wonder how much to read into that?

[ Boston Dynamics ]

You deserve to be free from endless chores! YOU! DESERVE! CHORE! FREEDOM!

Pretty sure this is teleoperated, so someone is still doing the chores, sadly.

[ MagicLab ]

[ Paper ] via [ EPFL LIS ]

[ Nissan ]

[ CERN ]

Congrats to Starship, the OG robotic delivery service, on their $90m raise.

[ Starship ]

[ GitHub ] via [ MIT ]

This is one of those things that’s far more difficult than it might look.

[ ROAM Lab ]

[ DEVANTHRO ]

It’s the world’s smallest humanoid robot, until someone comes out with slightly smaller servos!

[ Guinness ]

[ Deep Robotics ]

[ Harvard ]

[ DLR MIRO ]

[ Northwestern ]

It’s kind of astonishing how quadrotors have scaled over the past decade. Like, we’re now at the point where they’re verging on disposable, at least from a commercial or research perspective—for a bit over US $200, you can buy a little 27-gram, completely open-source drone, and all you have to do is teach it to fly. That’s where things do get a bit more challenging, though, because teaching drones to fly is not a straightforward process. Thanks to good simulation and techniques like reinforcement learning, it’s much easier to imbue drones with autonomy than it used to be. But it’s not typically a fast process, and it can be finicky to make a smooth transition from simulation to reality.

New York University’s Agile Robotics and Perception Lab has managed to streamline the process of getting basic autonomy to work on drones, and streamline it by a lot: The lab’s system is able to train a drone in simulation from nothing up to stable and controllable flying in 18 seconds flat on a MacBook Pro. And it actually takes longer to compile and flash the firmware onto the drone itself than it does for the entire training process.

ARPL NYU

So not only is the drone able to keep a stable hover while rejecting pokes and nudges and wind, but it’s also able to fly specific trajectories. Not bad for 18 seconds, right?

One of the things that typically slows down training times is the need to keep refining exactly what you’re training for, without refining it so much that you’re only training your system to fly in your specific simulation rather than the real world. The strategy used here is what the researchers call a curriculum (you can also think of it as a sort of lesson plan) to adjust the reward function used to train the system through reinforcement learning. The curriculum starts things off being more forgiving and gradually increasing the penalties to emphasize robustness and reliability. This is all about efficiency: Doing that training that you need to do in the way that it needs to be done to get the results you want, and no more.

There are other, more straightforward, tricks that optimize this technique for speed as well. The deep-reinforcement learning algorithms are particularly efficient, and leverage the hardware acceleration that comes along with Apple’s M-series processors. The simulator efficiency multiplies the benefits of the curriculum-driven sample efficiency of the reinforcement-learning pipeline, leading to that wicked-fast training time.

This approach isn’t limited to simple tiny drones—it’ll work on pretty much any drone, including bigger and more expensive ones, or even a drone that you yourself build from scratch.

Jonas Eschmann

We’re told that it took minutes rather than seconds to train a policy for the drone in the video above, although the researchers expect that 18 seconds is achievable even for a more complex drone like this in the near future. And it’s all open source, so you can, in fact, build a drone and teach it to fly with this system. But if you wait a little bit, it’s only going to get better: The researchers tell us that they’re working on integrating with the PX4 open source drone autopilot. Longer term, the idea is to have a single policy that can adapt to different environmental conditions, as well as different vehicle configurations, meaning that this could work on all kinds of flying robots rather than just quadrotors.

Everything you need to run this yourself is available on GitHub, and the paper is on ArXiv here.

About a decade ago, there was a lot of excitement in the robotics world around gecko-inspired directional adhesives, which are materials that stick without being sticky using the same van der Waals forces that allow geckos to scamper around on vertical panes of glass. They were used extensively in different sorts of climbing robots, some of them quite lovely. Gecko adhesives are uniquely able to stick to very smooth things where your only other option might be suction, which requires all kinds of extra infrastructure to work.

We haven’t seen gecko adhesives around as much of late, for a couple of reasons. First, the ability to only stick to smooth surfaces (which is what gecko adhesives are best at) is a bit of a limitation for mobile robots. And second, the gap between research and useful application is wide and deep and full of crocodiles. I’m talking about the mean kind of crocodiles, not the cuddly kind. But Flexiv Robotics has made gecko adhesives practical for robotic grasping in a commercial environment, thanks in part to a sort of robotic tongue that licks the gecko tape clean.

If you zoom way, way in on a gecko’s foot, you’ll see that each toe is covered in millions of hair-like nanostructures called setae. Each setae branches out at the end into hundreds of more hairs with flat bits at the end called spatulas. The result of this complex arrangement of setae and spatulas is that gecko toes have a ridiculous amount of surface area, meaning that they can leverage the extremely weak van der Waals forces between molecules to stick themselves to perfectly flat and smooth surfaces. This technique works exceptionally well: Geckos can hang from glass by a single toe, and a fully adhered gecko can hold something like 140 kg (which, unfortunately, seems to be an extrapolation rather than an experimental result). And luckily for the gecko, the structure of the spatulas makes the adhesion directional, so that when its toes are no longer being loaded, they can be easily peeled off of whatever they’re attached to.

Natural gecko adhesive structure, along with a synthetic adhesive (f).Gecko adhesion: evolutionary nanotechnology, by Kellar Autumn and Nick Gravish

Since geckos don’t “stick” to things in the sense that we typically use the word “sticky,” a better way of characterizing what geckos can do is as “dry adhesion,” as opposed to something that involves some sort of glue. You can also think about gecko toes as just being very, very high friction, and it’s this perspective that is particularly interesting in the context of robotic grippers.

This is Flexiv’s “Grav Enhanced” gripper, which uses a combination of pinch grasping and high friction gecko adhesive to lift heavy and delicate objects without having to squeeze them. When you think about a traditional robotic grasping system trying to lift something like a water balloon, you have to squeeze that balloon until the friction between the side of the gripper and the side of the balloon overcomes the weight of the balloon itself. The higher the friction, the lower the squeeze required, and although a water balloon might be an extreme example, maximizing gripper friction can make a huge difference when it comes to fragile or deformable objects.

There are a couple of problems with dry adhesive, however. The tiny structures that make the adhesive adhere can be prone to damage, and the fact that dry adhesive will stick to just about anything it can make good contact with means that it’ll rapidly accumulate dirt outside of a carefully controlled environment. In research contexts, these problems aren’t all that significant, but for a commercial system, you can’t have something that requires constant attention.

Flexiv says that the microstructure material that makes up their gecko adhesive was able to sustain two million gripping cycles without any visible degradation in performance, suggesting that as long as you use the stuff within the tolerances that it’s designed for, it should keep on adhering to things indefinitely—although trying to lift too much weight will tear the microstructures, ruining the adhesive properties after just a few cycles. And to keep the adhesive from getting clogged up with debris, Flexiv came up with this clever little cleaning station that acts like a little robotic tongue of sorts:

Interestingly, geckos themselves don’t seem to use their own tongues to clean their toes. They lick their eyeballs on the regular, like all normal humans do, but gecko toes appear to be self-cleaning, which is a pretty neat trick. It’s certainly possible to make self-cleaning synthetic gecko adhesive, but Flexiv tells us that “due to technical and practical limitations, replicating this process in our own gecko adhesive material is not possible. Essentially, we replicate the microstructure of a gecko’s footpad, but not its self-cleaning process.” This likely goes back to that whole thing about what works in a research context versus what works in a commercial context, and Flexiv needs their gecko adhesive to handle all those millions of cycles.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

Is “scamperiest” a word? If not, it should be, because this is the scamperiest robot I’ve ever seen.

[ ABS ]

[ GITAI ]

And if you want to get exhausted on behalf of a robot, the full 400-meter dash is below.

[ Hybrid Robotics ]

[ NASA ]

BigDog: 15 years of solving mobility problems its own way.

[ Boston Dynamics ]

[ Harvard SEAS ]

[ Paper ]

Thanks, Ram!

We are told that this new robot from HEBI is called “Mark Suckerberg” and that they’ve got a pretty cool application in mind for it, to be revealed later this year.

[ HEBI Robotics ]

Thanks, Dave!

[ SRL ETHZ ]

Carnegie Mellon researchers have built a system with a robotic arm atop a RangerMini 2.0 robotic cart from AgileX robotics to make what they’re calling a platform for “intelligent movement and processing.”

[ CMU ] via [ AgileX ]

[ Picassnake ]

[ NASA ]

Just because an object is around a corner doesn’t mean it has to be hidden. Non-line-of-sight imaging can peek around corners and spot those objects, but it has so far been limited to a narrow band of frequencies. Now, a new sensor can help extend this technique from working with visible light to infrared. This advance could help make autonomous vehicles safer, among other potential applications.

Non-line-of-sight imaging relies on the faint signals of light beams that have reflected off surfaces in order to reconstruct images. The ability to see around corners may prove useful for machine vision—for instance, helping autonomous vehicles foresee hidden dangers to better predict how to respond to them, says Xiaolong Hu, the senior author of the study and a professor at Tianjin University in Tianjin, China. It may also improve endoscopes that help doctors peer inside the body.

The light that non-line-of-sight imaging depends on is typically very dim, and until now, the detectors that were efficient and sensitive enough for non-line-of-sight imaging could only detect either visible or near-infrared light. Moving to longer wavelengths might have several advantages, such as dealing with less interference from sunshine, and the possibility of using lasers that are safe around eyes, Hu says.

Now Hu and his colleagues have for the first time performed non-line-of-sight imaging using 1,560- and 1,997-nanometer infrared wavelengths. “This extension in spectrum paves the way for more practical applications,” Hu says.

The researchers imaged several objects with a non-line-of-sight infrared camera, both without [middle column] and with [right column] de-noising algorithms.Tianjin University

In the new study, the researchers experimented with superconducting nanowire single-photon detectors. In each device, a 40-nanometer-wide niobium titanium nitride wire was cooled to about 2 kelvins (about –271 °C), rendering the wire superconductive. A single photon could disrupt this fragile state, generating electrical pulses that enabled the efficient detection of individual photons.

The scientists contorted the nanowire in each device into a fractal pattern that took on similar shapes at various magnifications. This let the sensor detect photons of all polarizations, boosting its efficiency.

The new detector was up to nearly three times as efficient as other single-photon detectors at sensing near- and mid-infrared light. This let the researchers perform non-line-of-sight imaging, achieving a spatial resolution of roughly 1.3 to 1.5 centimeters.

In addition to an algorithm that reconstructed non-line-of-sight images based off multiple scattered light rays, the scientists developed a new algorithm that helped remove noise from their data. When each pixel during the scanning process was given 5 milliseconds to collect photons, the new de-noising algorithm reduced the root mean square error—a measure of its deviation from a perfect image—of reconstructed images by about eightfold.

The researchers now plan to arrange multiple sensors into larger arrays to boost efficiency, reduce scanning time, and extend the distance over which imaging can take place, Hu says. They would also like to test their device in daylight conditions, he adds.

The scientists detailed their findings 30 November in the journal Optics Express.

Citing “no path to regulatory approval in the European Union,” Amazon and iRobot have announced the termination of an acquisition deal first announced in August of 2022 that would have made iRobot a part of Amazon and valued the robotics company at US $1.4 billion.

The European Commission released a statement today that explained some of its concerns, which to be fair, seem like reasonable things to be concerned about:

Amazon, for its part, characterizes this as “undue and disproportionate regulatory hurdles.” Whoever you believe is correct, the protracted strangulation of this acquisition deal has not been great for iRobot, and its termination is potentially disastrous—Amazon will have to pay iRobot a $94 million termination fee, which is basically nothing for it, and meanwhile iRobot is already laying off 350 people, or 31 percent of its head count.

From one of iRobot’s press releases:

The reason that I don’t feel much better after reading that statement is that Colin Angle has already stepped down as chairman and CEO of iRobot. Angle was one of the founders of iRobot (along with Rodney Brooks and Helen Greiner) and has stuck with the company for its entire 30+ year existence, until just now. So, that’s not great. Also, I’m honestly not sure how iRobot is going to create much in the way of home innovations since the press release states that the company is “pausing all work related to non-floor care innovations, including air purification, robotic lawn mowing and education,” while also “reducing R&D expense by approximately $20 million year-over-year.”

iRobot’s lawn mower has been paused for a while now, so it’s not a huge surprise that nothing will move forward there, but a pause on the education robots like Create and Root is a real blow to the robotics community. And even if iRobot is focusing on floor-care innovations, I’m not sure how much innovation will be possible with a slashed R&D budget amidst huge layoffs.

Sigh.

On LinkedIn, Colin Angle wrote a little bit about what he called “the magic of iRobot”:

And that’s what I personally find so worrying about all of this. iRobot has a treasured history of innovation which is full of successes and failures and really weird stuff, and it’s hard to see how that will be able to effectively continue. Here are a couple of my favorite weird iRobot things, including a PackBot that flies (for a little bit) and a morphing blobular robot:

I suppose it’s worth pointing out that the weirdest stuff (like in the videos above) is all over a decade old, and you can reasonably ask whether iRobot was that kind of company anymore even before this whole Amazon thing happened. The answer is probably not, since the company has chosen to focus almost exclusively on floor-care robots. But even there we’ve seen consistent innovation in hardware and software that pretty much every floor-care robot company seems to then pick up on about a year later. This is not to say that other floor-care robots can’t innovate, but it’s undeniable that iRobot has been a driving force behind that industry. Will that continue? I really hope so.

Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.

Enjoy today’s videos!

If this seems a little bit unsettling, that’s intentional! The project was designed to explore the concepts of trust and risk in the context of robots, and of using technology to influence emotion.

[ medusai ] via [ Georgia Tech ]

Thanks, Gil!

[ NASA JPL ]

Sometimes all it takes is one good punch, and then you can just collapse.

[ Paper ] via [ IHMC ]

Thanks, Robert!

[ Dusty Robotics ]

[ DreamRiser ]

The work that Henry Evans is doing with Stretch (along with Hello Robot and Maya Cakmak’s lab at the University of Washington) will be presented at Humanoids this spring.

[ UW HCRL ]

Thanks, Stefan!

I like to imagine that these are just excerpts from one very long walk that Digit took around San Francisco.

[ Hybrid Robotics Lab ]

[ LimX Dynamics ]

[ OpenDR ] via [ PAL Robotics ]

[ ARCHES ]

Christoph Bartneck gives a talk entitled “Social robots: The end of the beginning or the beginning of the end?”

[ Christoph Bartneck ]

[ Berkeley ]