Tag Archives: learns

#437769 Q&A: Facebook’s CTO Is at War With ...

Posted on November 29, 2020 by Android

Photo: Patricia de Melo Moreira/AFP/Getty Images

Facebook chief technology officer Mike Schroepfer leads the company’s AI and integrity efforts.

Facebook’s challenge is huge. Billions of pieces of content—short and long posts, images, and combinations of the two—are uploaded to the site daily from around the world. And any tiny piece of that—any phrase, image, or video—could contain so-called bad content.

In its early days, Facebook relied on simple computer filters to identify potentially problematic posts by their words, such as those containing profanity. These automatically filtered posts, as well as posts flagged by users as offensive, went to humans for adjudication.

In 2015, Facebook started using artificial intelligence to cull images that contained nudity, illegal goods, and other prohibited content; those images identified as possibly problematic were sent to humans for further review.

By 2016, more offensive photos were reported by Facebook’s AI systems than by Facebook users (and that is still the case).

In 2018, Facebook CEO Mark Zuckerberg made a bold proclamation: He predicted that within five or ten years, Facebook’s AI would not only look for profanity, nudity, and other obvious violations of Facebook’s policies. The tools would also be able to spot bullying, hate speech, and other misuse of the platform, and put an immediate end to them.

Today, automated systems using algorithms developed with AI scan every piece of content between the time when a user completes a post and when it is visible to others on the site—just fractions of a second. In most cases, a violation of Facebook’s standards is clear, and the AI system automatically blocks the post. In other cases, the post goes to human reviewers for a final decision, a workforce that includes 15,000 content reviewers and another 20,000 employees focused on safety and security, operating out of more than 20 facilities around the world.

In the first quarter of this year, Facebook removed or took other action (like appending a warning label) on more than 9.6 million posts involving hate speech, 8.6 million involving child nudity or exploitation, almost 8 million posts involving the sale of drugs, 2.3 million posts involving bullying and harassment, and tens of millions of posts violating other Facebook rules.

Right now, Facebook has more than 1,000 engineers working on further developing and implementing what the company calls “integrity” tools. Using these systems to screen every post that goes up on Facebook, and doing so in milliseconds, is sucking up computing resources. Facebook chief technology officer Mike Schroepfer, who is heading up Facebook’s AI and integrity efforts, spoke with IEEE Spectrum about the team’s progress on building an AI system that detects bad content.

Since that discussion, Facebook’s policies around hate speech have come under increasing scrutiny, with particular attention on divisive posts by political figures. A group of major advertisers in June announced that they would stop advertising on the platform while reviewing the situation, and civil rights groups are putting pressure on others to follow suit until Facebook makes policy changes related to hate speech and groups that promote hate, misinformation, and conspiracies.

Facebook CEO Mark Zuckerberg responded with news that Facebook will widen the category of what it considers hateful content in ads. Now the company prohibits claims that people from a specific race, ethnicity, national origin, religious affiliation, caste, sexual orientation, gender identity, or immigration status are a threat to the physical safety, health, or survival of others. The policy change also aims to better protect immigrants, migrants, refugees, and asylum seekers from ads suggesting these groups are inferior or expressing contempt. Finally, Zuckerberg announced that the company will label some problematic posts by politicians and government officials as content that violates Facebook’s policies.

However, civil rights groups say that’s not enough. And an independent audit released in July also said that Facebook needs to go much further in addressing civil rights concerns and disinformation.

Schroepfer indicated that Facebook’s AI systems are designed to quickly adapt to changes in policy. “I don’t expect considerable technical changes are needed to adjust,” he told Spectrum.

This interview has been edited and condensed for clarity.

IEEE Spectrum: What are the stakes of content moderation? Is this an existential threat to Facebook? And is it critical that you deal well with the issue of election interference this year?

Schroepfer: It’s probably existential; it’s certainly massive. We are devoting a tremendous amount of our attention to it.

The idea that anyone could meddle in an election is deeply disturbing and offensive to all of us here, just as people and citizens of democracies. We don’t want to see that happen anywhere, and certainly not on our watch. So whether it’s important to the company or not, it’s important to us as people. And I feel a similar way on the content-moderation side.

There are not a lot of easy choices here. The only way to prevent people, with certainty, from posting bad things is to not let them post anything. We can take away all voice and just say, “Sorry, the Internet’s too dangerous. No one can use it.” That will certainly get rid of all hate speech online. But I don’t want to end up in that world. And there are variants of that world that various governments are trying to implement, where they get to decide what’s true or not, and you as a person don’t. I don’t want to get there either.

My hope is that we can build a set of tools that make it practical for us to do a good enough job, so that everyone is still excited about the idea that anyone can share what they want, and so that Facebook is a safe and reasonable place for people to operate in.

Spectrum: You joined Facebook in 2008, before AI was part of the company’s toolbox. When did that change? When did you begin to think that AI tools would be useful to Facebook?

Schroepfer: Ten years ago, AI wasn’t commercially practical; the technology just didn’t work very well. In 2012, there was one of those moments that a lot of people point to as the beginning of the current revolution in deep learning and AI. A computer-vision model—a neural network—was trained using what we call supervised training, and it turned out to be better than all the existing models.

Spectrum: How is that training done, and how did computer-vision models come to Facebook?

Image: Facebook

Just Broccoli? Facebook’s image analysis algorithms can tell the difference between marijuana [left] and tempura broccoli [right] better than some humans.

Schroepfer: Say I take a bunch of photos and I have people look at them. If they see a photo of a cat, they put a text label that says cat; if it’s one of a dog, the text label says dog. If you build a big enough data set and feed that to the neural net, it learns how to tell the difference between cats and dogs.

Prior to 2012, it didn’t work very well. And then in 2012, there was this moment where it seemed like, “Oh wow, this technique might work.” And a few years later we were deploying that form of technology to help us detect problematic imagery.

Spectrum: Do your AI systems work equally well on all types of prohibited content?

Schroepfer: Nudity was technically easiest. I don’t need to understand language or culture to understand that this is either a naked human or not. Violence is a much more nuanced problem, so it was harder technically to get it right. And with hate speech, not only do you have to understand the language, it may be very contextual, even tied to recent events. A week before the Christchurch shooting [New Zealand, 2019], saying “I wish you were in the mosque” probably doesn’t mean anything. A week after, that might be a terrible thing to say.

Spectrum: How much progress have you made on hate speech?

Schroepfer: AI, in the first quarter of 2020, proactively detected 88.8 percent of the hate-speech content we removed, up from 80.2 percent in the previous quarter. In the first quarter of 2020, we took action on 9.6 million pieces of content for violating our hate-speech policies.

Image: Facebook

Off Label: Sometimes image analysis isn’t enough to determine whether a picture posted violates the company’s policies. In considering these candy-colored vials of marijuana, for example, the algorithms can look at any accompanying text and, if necessary, comments on the post.

Spectrum: It sounds like you’ve expanded beyond tools that analyze images and are also using AI tools that analyze text.

Schroepfer: AI started off as very siloed. People worked on language, people worked on computer vision, people worked on video. We’ve put these things together—in production, not just as research—into multimodal classifiers.

[Schroepfer shows a photo of a pan of Rice Krispies treats, with text referring to it as a “potent batch”] This is a case in which you have an image, and then you have the text on the post. This looks like Rice Krispies. On its own, this image is fine. You put the text together with it in a bigger model; that can then understand what’s going on. That didn’t work five years ago.

Spectrum: Today, every post that goes up on Facebook is immediately checked by automated systems. Can you explain that process?

Image: Facebook

Bigger Picture: Identifying hate speech is often a matter of context. Either the text or the photo in this post isn’t hateful standing alone, but putting them together tells a different story.

Schroepfer: You upload an image and you write some text underneath it, and the systems look at both the image and the text to try to see which, if any, policies it violates. Those decisions are based on our Community Standards. It will also look at other signals on the posts, like the comments people make.

It happens relatively instantly, though there may be times things happen after the fact. Maybe you uploaded a post that had misinformation in it, and at the time you uploaded it, we didn’t know it was misinformation. The next day we fact-check something and scan again; we may find your post and take it down. As we learn new things, we’re going to go back through and look for violations of what we now know to be a problem. Or, as people comment on your post, we might update our understanding of it. If people are saying, “That’s terrible,” or “That’s mean,” or “That looks fake,” those comments may be an interesting signal.

Spectrum: How is Facebook applying its AI tools to the problem of election interference?

Schroepfer: I would split election interference into two categories. There are times when you’re going after the content, and there are times you’re going after the behavior or the authenticity of the person.

On content, if you’re sharing misinformation, saying, “It’s super Wednesday, not super Tuesday, come vote on Wednesday,” that’s a problem whether you’re an American sitting in California or a foreign actor.

Other times, people create a series of Facebook pages pretending they’re Americans, but they’re really a foreign entity. That is a problem on its own, even if all the content they’re sharing completely meets our Community Standards. The problem there is that you have a foreign government running an information operation.

There, you need different tools. What you’re trying to do is put pieces together, to say, “Wait a second. All of these pages—Martians for Justice, Moonlings for Justice, and Venusians for Justice”—are all run by an administrator with an IP address that’s outside the United States. So they’re all connected, even though they’re pretending to not be connected. That’s a very different problem than me sitting in my office in Menlo Park [Calif.] sharing misinformation.

I’m not going to go into lots of technical detail, because this is an area of adversarial nature. The fundamental problem you’re trying to solve is that there’s one entity coordinating the activity of a bunch of things that look like they’re not all one thing. So this is a series of Instagram accounts, or a series of Facebook pages, or a series of WhatsApp accounts, and they’re pretending to be totally different things. We’re looking for signals that these things are related in some way. And we’re looking through the graph [what Facebook calls its map of relationships between users] to understand the properties of this network.

Spectrum: What cutting-edge AI tools and methods have you been working on lately?

Schroepfer: Supervised learning, with humans setting up the instruction process for the AI systems, is amazingly effective. But it has a very obvious flaw: the speed at which you can develop these things is limited by how fast you can curate the data sets. If you’re dealing in a problem domain where things change rapidly, you have to rebuild a new data set and retrain the whole thing.

Self-supervision is inspired by the way people learn, by the way kids explore the world around them. To get computers to do it themselves, we take a bunch of raw data and build a way for the computer to construct its own tests. For language, you scan a bunch of Web pages, and the computer builds a test where it takes a sentence, eliminates one of the words, and figures out how to predict what word belongs there. And because it created the test, it actually knows the answer. I can use as much raw text as I can find and store because it’s processing everything itself and doesn’t require us to sit down and build the information set. In the last two years there has been a revolution in language understanding as a result of AI self-supervised learning.

Spectrum: What else are you excited about?

Schroepfer: What we’ve been working on over the last few years is multilingual understanding. Usually, when I’m trying to figure out, say, whether something is hate speech or not I have to go through the whole process of training the model in every language. I have to do that one time for every language. When you make a post, the first thing we have to figure out is what language your post is in. “Ah, that’s Spanish. So send it to the Spanish hate-speech model.”

We’ve started to build a multilingual model—one box where you can feed in text in 40 different languages and it determines whether it’s hate speech or not. This is way more effective and easier to deploy.

To geek out for a second, just the idea that you can build a model that understands a concept in multiple languages at once is crazy cool. And it not only works for hate speech, it works for a variety of things.

When we started working on this multilingual model years ago, it performed worse than every single individual model. Now, it not only works as well as the English model, but when you get to the languages where you don’t have enough data, it’s so much better. This rapid progress is very exciting.

Spectrum: How do you move new AI tools from your research labs into operational use?

Schroepfer: Engineers trying to make the next breakthrough will often say, “Cool, I’ve got a new thing and it achieved state-of-the-art results on machine translation.” And we say, “Great. How long does it take to run in production?” They say, “Well, it takes 10 seconds for every sentence to run on a CPU.” And we say, “It’ll eat our whole data center if we deploy that.” So we take that state-of-the-art model and we make it 10 or a hundred or a thousand times more efficient, maybe at the cost of a little bit of accuracy. So it’s not as good as the state-of-the-art version, but it’s something we can actually put into our data centers and run in production.

Spectrum: What’s the role of the humans in the loop? Is it true that Facebook currently employs 35,000 moderators?

Schroepfer: Yes. Right now our goal is not to reduce that. Our goal is to do a better job catching bad content. People often think that the end state will be a fully automated system. I don’t see that world coming anytime soon.

As automated systems get more sophisticated, they take more and more of the grunt work away, freeing up the humans to work on the really gnarly stuff where you have to spend an hour researching.

We also use AI to give our human moderators power tools. Say I spot this new meme that is telling everyone to vote on Wednesday rather than Tuesday. I have a tool in front of me that says, “Find variants of that throughout the system. Find every photo with the same text, find every video that mentions this thing and kill it in one shot.” Rather than, I found this one picture, but then a bunch of other people upload that misinformation in different forms.

Another important aspect of AI is that anything I can do to prevent a person from having to look at terrible things is time well spent. Whether it’s a person employed by us as a moderator or a user of our services, looking at these things is a terrible experience. If I can build systems that take the worst of the worst, the really graphic violence, and deal with that in an automated fashion, that’s worth a lot to me. Continue reading →

Posted in Human Robots

#437624 AI-Powered Drone Learns Extreme ...

Posted on November 22, 2020 by Android

Quadrotors are among the most agile and dynamic machines ever created. In the hands of a skilled human pilot, they can do some astonishing series of maneuvers. And while autonomous flying robots have been getting better at flying dynamically in real-world environments, they still haven’t demonstrated the same level of agility of manually piloted ones.

Now researchers from the Robotics and Perception Group at the University of Zurich and ETH Zurich, in collaboration with Intel, have developed a neural network training method that “enables an autonomous quadrotor to fly extreme acrobatic maneuvers with only onboard sensing and computation.” Extreme.

There are two notable things here: First, the quadrotor can do these extreme acrobatics outdoors without any kind of external camera or motion-tracking system to help it out (all sensing and computing is onboard). Second, all of the AI training is done in simulation, without the need for an additional simulation-to-real-world (what researchers call “sim-to-real”) transfer step. Usually, a sim-to-real transfer step means putting your quadrotor into one of those aforementioned external tracking systems, so that it doesn’t completely bork itself while trying to reconcile the differences between the simulated world and the real world, where, as the researchers wrote in a paper describing their system, “even tiny mistakes can result in catastrophic outcomes.”

To enable “zero-shot” sim-to-real transfer, the neural net training in simulation uses an expert controller that knows exactly what’s going on to teach a “student controller” that has much less perfect knowledge. That is, the simulated sensory input that the student ends up using as it learns to follow the expert has been abstracted to present the kind of imperfect, imprecise data it’s going to encounter in the real world. This can involve things like abstracting away the image part of the simulation until you’d have no way of telling the difference between abstracted simulation and abstracted reality, which is what allows the system to make that sim-to-real leap.

The simulation environment that the researchers used was Gazebo, slightly modified to better simulate quadrotor physics. Meanwhile, over in reality, a custom 1.5-kilogram quadrotor with a 4:1 thrust to weight ratio performed the physical experiments, using only a Nvidia Jetson TX2 computing board and an Intel RealSense T265, a dual fisheye camera module optimized for V-SLAM. To challenge the learning system, it was trained to perform three acrobatic maneuvers plus a combo of all of them:

Image: University of Zurich/ETH Zurich/Intel

Reference trajectories for acrobatic maneuvers. Top row, from left: Power Loop, Barrel Roll, and Matty Flip. Bottom row: Combo.

All of these maneuvers require high accelerations of up to 3 g’s and careful control, and the Matty Flip is particularly challenging, at least for humans, because the whole thing is done while the drone is flying backwards. Still, after just a few hours of training in simulation, the drone was totally real-world competent at these tricks, and could even extrapolate a little bit to perform maneuvers that it was not explicitly trained on, like doing multiple loops in a row. Where humans still have the advantage over drones is (as you might expect since we’re talking about robots) is quickly reacting to novel or unexpected situations. And when you’re doing this sort of thing outdoors, novel and unexpected situations are everywhere, from a gust of wind to a jealous bird.

For more details, we spoke with Antonio Loquercio from the University of Zurich’s Robotics and Perception Group.

IEEE Spectrum: Can you explain how the abstraction layer interfaces with the simulated sensors to enable effective sim-to-real transfer?

Antonio Loquercio: The abstraction layer applies a specific function to the raw sensor information. Exactly the same function is applied to the real and simulated sensors. The result of the function, which is “abstracted sensor measurements,” makes simulated and real observation of the same scene similar. For example, suppose we have a sequence of simulated and real images. We can very easily tell apart the real from the simulated ones given the difference in rendering. But if we apply the abstraction function of “feature tracks,” which are point correspondences in time, it becomes very difficult to tell which are the simulated and real feature tracks, since point correspondences are independent of the rendering. This applies for humans as well as for neural networks: Training policies on raw images gives low sim-to-real transfer (since images are too different between domains), while training on the abstracted images has high transfer abilities.

How useful is visual input from a camera like the Intel RealSense T265 for state estimation during such aggressive maneuvers? Would using an event camera substantially improve state estimation?

Our end-to-end controller does not require a state estimation module. It shares however some components with traditional state estimation pipelines, specifically the feature extractor and the inertial measurement unit (IMU) pre-processing and integration function. The input of the neural networks are feature tracks and integrated IMU measurements. When looking at images with low features (for example when the camera points to the sky), the neural net will mainly rely on IMU. When more features are available, the network uses to correct the accumulated drift from IMU. Overall, we noticed that for very short maneuvers IMU measurements were sufficient for the task. However, for longer ones, visual information was necessary to successfully address the IMU drift and complete the maneuver. Indeed, visual information reduces the odds of a crash by up to 30 percent in the longest maneuvers. We definitely think that event camera can improve even more the current approach since they could provide valuable visual information during high speed.

“The Matty Flip is probably one of the maneuvers that our approach can do very well … It is super challenging for humans, since they don’t see where they’re going and have problems in estimating their speed. For our approach the maneuver is no problem at all, since we can estimate forward velocities as well as backward velocities.”
—Antonio Loquercio, University of Zurich

You describe being able to train on “maneuvers that stretch the abilities of even expert human pilots.” What are some examples of acrobatics that your drones might be able to do that most human pilots would not be capable of?

The Matty Flip is probably one of the maneuvers that our approach can do very well, but human pilots find very challenging. It basically entails doing a high speed power loop by always looking backward. It is super challenging for humans, since they don’t see where they’re going and have problems in estimating their speed. For our approach the maneuver is no problem at all, since we can estimate forward velocities as well as backward velocities.

What are the limits to the performance of this system?

At the moment the main limitation is the maneuver duration. We never trained a controller that could perform maneuvers longer than 20 seconds. In the future, we plan to address this limitation and train general controllers which can fly in that agile way for significantly longer with relatively small drift. In this way, we could start being competitive against human pilots in drone racing competitions.

Can you talk about how the techniques developed here could be applied beyond drone acrobatics?

The current approach allows us to do acrobatics and agile flight in free space. We are now working to perform agile flight in cluttered environments, which requires a higher degree of understanding of the surrounding with respect to this project. Drone acrobatics is of course only an example application. We selected it because it makes a stress test of the controller performance. However, several other applications which require fast and agile flight can benefit from our approach. Examples are delivery (we want our Amazon packets always faster, don’t we?), search and rescue, or inspection. Going faster allows us to cover more space in less time, saving battery costs. Indeed, agile flight has very similar battery consumption of slow hovering for an autonomous drone.

“Deep Drone Acrobatics,” by Elia Kaufmann, Antonio Loquercio, René Ranftl, Matthias Müller, Vladlen Koltun, and Davide Scaramuzza from the Robotics and Perception Group at the University of Zurich and ETH Zurich, and Intel’s Intelligent Systems Lab, was presented at RSS 2020. Continue reading →

Posted in Human Robots

#437614 Video Friday: Poimo Is a Portable ...

Posted on November 21, 2020 by Android

Video Friday is your weekly selection of awesome robotics videos, collected by your Automaton bloggers. We’ll also be posting a weekly calendar of upcoming robotics events for the next few months; here's what we have so far (send us your events!):

IROS 2020 – October 25-29, 2020 – [Online]
ROS World 2020 – November 12, 2020 – [Online]
CYBATHLON 2020 – November 13-14, 2020 – [Online]
ICSR 2020 – November 14-16, 2020 – Golden, Colo., USA
Let us know if you have suggestions for next week, and enjoy today's videos.

Engineers at the University of California San Diego have built a squid-like robot that can swim untethered, propelling itself by generating jets of water. The robot carries its own power source inside its body. It can also carry a sensor, such as a camera, for underwater exploration.

[ UCSD ]

Thanks Ioana!

Shark Robotics, French and European leader in Unmanned Ground Vehicles, is announcing today a disinfection add-on for Boston Dynamics Spot robot, designed to fight the COVID-19 pandemic. The Spot robot with Shark’s purpose-built disinfection payload can decontaminate up to 2,000 m2 in 15 minutes, in any space that needs to be sanitized – such as hospitals, metro stations, offices, warehouses or facilities.

[ Shark Robotics ]

Here’s an update on the Poimo portable inflatable mobility project we wrote about a little while ago; while not strictly robotics, it seems like it holds some promise for rapidly developing different soft structures that robotics might find useful.

[ University of Tokyo ]

Thanks Ryuma!

Pretty cool that you can do useful force feedback teleop while video chatting through a “regular broadband Internet connection.” Although, what “regular” means to you is a bit subjective, right?

[ HEBI Robotics ]

Thanks Dave!

While NASA's Mars rover Perseverance travels through space toward the Red Planet, its nearly identical rover twin is hard at work on Earth. The vehicle system test bed (VSTB) rover named OPTIMISM is a full-scale engineering version of the Mars-bound rover. It is used to test hardware and software before the commands are sent up to the Perseverance rover.

[ NASA ]

Jacquard takes ordinary, familiar objects and enhances them with new digital abilities and experiences, while remaining true to their original purpose — like being your favorite jacket, backpack or a pair of shoes that you love to wear.

Our ambition is simple: to make life easier. By staying connected to your digital world, your things can do so much more. Skip a song by brushing your sleeve. Take a picture by tapping on a shoulder strap. Get reminded about the phone you left behind with a blink of light or a haptic buzz on your cuff.

[ Google ATAP ]

Should you attend the IROS 2020 workshop on “Planetary Exploration Robots: Challenges and Opportunities”? Of course you should!

[ Workshop ]

Kuka makes a lot of these videos where I can’t help but think that if they put as much effort into programming the robot as they did into producing the video, the result would be much more impressive.

[ Kuka ]

The Colorado School of Mines is one of the first customers to buy a Spot robot from Boston Dynamics to help with robotics research. Watch as scientists take Spot into the school's mine for the first time.

[ HCR ] via [ CNET ]

A very interesting soft(ish) actuator from Ayato Kanada at Kyushu University's Control Engineering Lab.

A flexible ultrasonic motor (FUSM), which generates linear motion as a novel soft actuator. This motor consists of a single metal cube stator with a hole and an elastic elongated coil spring inserted into the hole. When voltages are applied to piezoelectric plates on the stator, the coil spring moves back and forward as a linear slider. In the FUSM that uses the friction drive as the principle, the most important parameter for optimizing its output is the preload between the stator and slider. The coil spring has a slightly larger diameter than the stator hole and generates the preload by expanding in a radial direction. The coil springs act not only as a flexible slider but also as a resistive positional sensor. Changes in the resistance between the stator and the coil spring end are converted to a voltage and used for position detection.

[ Control Engineering Lab ]

Thanks Ayato!

We show how to use the limbs of a quadruped robot to identify fine-grained soil, representative for Martian regolith.

[ Paper ] via [ ANYmal Research ]

PR2 is serving breakfast and cleaning up afterwards. It’s slow, but all you have to do is eat and leave.

That poor PR2 is a little more naked than it's probably comfortable with.

[ EASE ]

NVIDIA researchers present a hierarchical framework that combines model-based control and reinforcement learning (RL) to synthesize robust controllers for a quadruped robot (the Unitree Laikago).

[ NVIDIA ]

What's interesting about this assembly task is that the robot is using its arm only for positioning, and doing the actual assembly with just fingers.

[ RC2L ]

In this electronics assembly application, Kawasaki's cobot duAro2 uses a tool changing station to tackle a multitude of tasks and assemble different CPU models.

Okay but can it apply thermal paste to a CPU in the right way? Personally, I find that impossible.

[ Kawasaki ]

You only need to watch this video long enough to appreciate the concept of putting a robot on a robot.

[ Impress ]

In this lecture, we’ll hear from the man behind one of the biggest robotics companies in the world, Boston Dynamics, whose robotic dog, Spot, has been used to encourage social distancing in Singapore and is now getting ready for FDA approval to be able to measure patients’ vital signs in hospitals.

[ Alan Turing Institute ]

Greg Kahn from UC Berkeley wrote in to share his recent dissertation talk on “Mobile Robot Learning.”

In order to create mobile robots that can autonomously navigate real-world environments, we need generalizable perception and control systems that can reason about the outcomes of navigational decisions. Learning-based methods, in which the robot learns to navigate by observing the outcomes of navigational decisions in the real world, offer considerable promise for obtaining these intelligent navigation systems. However, there are many challenges impeding mobile robots from autonomously learning to act in the real-world, in particular (1) sample-efficiency–how to learn using a limited amount of data? (2) supervision–how to tell the robot what to do? and (3) safety–how to ensure the robot and environment are not damaged or destroyed during learning? In this talk, I will present deep reinforcement learning methods for addressing these real world mobile robot learning challenges and show results which enable ground and aerial robots to navigate in complex indoor and outdoor environments.

[ UC Berkeley ]

Thanks Greg!

Leila Takayama from UC Santa Cruz (and previously Google X and Willow Garage) gives a talk entitled “Toward a more human-centered future of robotics.”

Robots are no longer only in outer space, in factory cages, or in our imaginations. We interact with robotic agents when withdrawing cash from bank ATMs, driving cars with adaptive cruise control, and tuning our smart home thermostats. In the moment of those interactions with robotic agents, we behave in ways that do not necessarily align with the rational belief that robots are just plain machines. Through a combination of controlled experiments and field studies, we use theories and concepts from the social sciences to explore ways that human and robotic agents come together, including how people interact with personal robots and how people interact through telepresence robots. Together, we will explore topics and raise questions about the psychology of human-robot interaction and how we could invent a future of a more human-centered robotics that we actually want to live in.

[ Leila Takayama ]

Roboticist and stand-up comedian Naomi Fitter from Oregon State University gives a talk on “Everything I Know about Telepresence.”

Telepresence robots hold promise to connect people by providing videoconferencing and navigation abilities in far-away environments. At the same time, the impacts of current commercial telepresence robots are not well understood, and circumstances of robot use including internet connection stability, odd personalizations, and interpersonal relationship between a robot operator and people co-located with the robot can overshadow the benefit of the robot itself. And although the idea of telepresence robots has been around for over two decades, available nonverbal expressive abilities through telepresence robots are limited, and suitable operator user interfaces for the robot (for example, controls that allow for the operator to hold a conversation and move the robot simultaneously) remain elusive. So where should we be using telepresence robots? Are there any pitfalls to watch out for? What do we know about potential robot expressivity and user interfaces? This talk will cover my attempts to address these questions and ways in which the robotics research community can build off of this work

[ Talking Robotics ] Continue reading →

Posted in Human Robots

#437575 AI-Directed Robotic Hand Learns How to ...

Posted on November 19, 2020 by Android

Reaching for a nearby object seems like a mindless task, but the action requires a sophisticated neural network that took humans millions of years to evolve. Now, robots are acquiring that same ability using artificial neural networks. In a recent study, a robotic hand “learns” to pick up objects of different shapes and hardness using three different grasping motions.

The key to this development is something called a spiking neuron. Like real neurons in the brain, artificial neurons in a spiking neural network (SNN) fire together to encode and process temporal information. Researchers study SNNs because this approach may yield insights into how biological neural networks function, including our own.

“The programming of humanoid or bio-inspired robots is complex,” says Juan Camilo Vasquez Tieck, a research scientist at FZI Forschungszentrum Informatik in Karlsruhe, Germany. “And classical robotics programming methods are not always suitable to take advantage of their capabilities.”

Conventional robotic systems must perform extensive calculations, Tieck says, to track trajectories and grasp objects. But a robotic system like Tieck’s, which relies on a SNN, first trains its neural net to better model system and object motions. After which it grasps items more autonomously—by adapting to the motion in real-time.

The new robotic system by Tieck and his colleagues uses an existing robotic hand, called a Schunk SVH 5-finger hand, which has the same number of fingers and joints as a human hand.

The researchers incorporated a SNN into their system, which is divided into several sub-networks. One sub-network controls each finger individually, either flexing or extending the finger. Another concerns each type of grasping movement, for example whether the robotic hand will need to do a pinching, spherical or cylindrical movement.

For each finger, a neural circuit detects contact with an object using the currents of the motors and the velocity of the joints. When contact with an object is detected, a controller is activated to regulate how much force the finger exerts.

“This way, the movements of generic grasping motions are adapted to objects with different shapes, stiffness and sizes,” says Tieck. The system can also adapt its grasping motion quickly if the object moves or deforms.

The robotic grasping system is described in a study published October 24 in IEEE Robotics and Automation Letters. The researchers’ robotic hand used its three different grasping motions on objects without knowing their properties. Target objects included a plastic bottle, a soft ball, a tennis ball, a sponge, a rubber duck, different balloons, a pen, and a tissue pack. The researchers found, for one, that pinching motions required more precision than cylindrical or spherical grasping motions.

“For this approach, the next step is to incorporate visual information from event-based cameras and integrate arm motion with SNNs,” says Tieck. “Additionally, we would like to extend the hand with haptic sensors.”

The long-term goal, he says, is to develop “a system that can perform grasping similar to humans, without intensive planning for contact points or intense stability analysis, and [that is] able to adapt to different objects using visual and haptic feedback.” Continue reading →

Posted in Human Robots

#436911 Scientists Linked Artificial and ...

Posted on March 11, 2020 by Android

Scientists have linked up two silicon-based artificial neurons with a biological one across multiple countries into a fully-functional network. Using standard internet protocols, they established a chain of communication whereby an artificial neuron controls a living, biological one, and passes on the info to another artificial one.

Whoa.

We’ve talked plenty about brain-computer interfaces and novel computer chips that resemble the brain. We’ve covered how those “neuromorphic” chips could link up into tremendously powerful computing entities, using engineered communication nodes called artificial synapses.

As Moore’s law is dying, we even said that neuromorphic computing is one path towards the future of extremely powerful, low energy consumption artificial neural network-based computing—in hardware—that could in theory better link up with the brain. Because the chips “speak” the brain’s language, in theory they could become neuroprosthesis hubs far more advanced and “natural” than anything currently possible.

This month, an international team put all of those ingredients together, turning theory into reality.

The three labs, scattered across Padova, Italy, Zurich, Switzerland, and Southampton, England, collaborated to create a fully self-controlled, hybrid artificial-biological neural network that communicated using biological principles, but over the internet.

The three-neuron network, linked through artificial synapses that emulate the real thing, was able to reproduce a classic neuroscience experiment that’s considered the basis of learning and memory in the brain. In other words, artificial neuron and synapse “chips” have progressed to the point where they can actually use a biological neuron intermediary to form a circuit that, at least partially, behaves like the real thing.

That’s not to say cyborg brains are coming soon. The simulation only recreated a small network that supports excitatory transmission in the hippocampus—a critical region that supports memory—and most brain functions require enormous cross-talk between numerous neurons and circuits. Nevertheless, the study is a jaw-dropping demonstration of how far we’ve come in recreating biological neurons and synapses in artificial hardware.

And perhaps one day, the currently “experimental” neuromorphic hardware will be integrated into broken biological neural circuits as bridges to restore movement, memory, personality, and even a sense of self.

The Artificial Brain Boom
One important thing: this study relies heavily on a decade of research into neuromorphic computing, or the implementation of brain functions inside computer chips.

The best-known example is perhaps IBM’s TrueNorth, which leveraged the brain’s computational principles to build a completely different computer than what we have today. Today’s computers run on a von Neumann architecture, in which memory and processing modules are physically separate. In contrast, the brain’s computing and memory are simultaneously achieved at synapses, small “hubs” on individual neurons that talk to adjacent ones.

Because memory and processing occur on the same site, biological neurons don’t have to shuttle data back and forth between processing and storage compartments, massively reducing processing time and energy use. What’s more, a neuron’s history will also influence how it behaves in the future, increasing flexibility and adaptability compared to computers. With the rise of deep learning, which loosely mimics neural processing as the prima donna of AI, the need to reduce power while boosting speed and flexible learning is becoming ever more tantamount in the AI community.

Neuromorphic computing was partially born out of this need. Most chips utilize special ingredients that change their resistance (or other physical characteristics) to mimic how a neuron might adapt to stimulation. Some chips emulate a whole neuron, that is, how it responds to a history of stimulation—does it get easier or harder to fire? Others imitate synapses themselves, that is, how easily they will pass on the information to another neuron.

Although single neuromorphic chips have proven to be far more efficient and powerful than current computer chips running machine learning algorithms in toy problems, so far few people have tried putting the artificial components together with biological ones in the ultimate test.

That’s what this study did.

A Hybrid Network
Still with me? Let’s talk network.

It’s gonna sound complicated, but remember: learning is the formation of neural networks, and neurons that fire together wire together. To rephrase: when learning, neurons will spontaneously organize into networks so that future instances will re-trigger the entire network. To “wire” together, downstream neurons will become more responsive to their upstream neural partners, so that even a whisper will cause them to activate. In contrast, some types of stimulation will cause the downstream neuron to “chill out” so that only an upstream “shout” will trigger downstream activation.

Both these properties—easier or harder to activate downstream neurons—are essentially how the brain forms connections. The “amping up,” in neuroscience jargon, is long-term potentiation (LTP), whereas the down-tuning is LTD (long-term depression). These two phenomena were first discovered in the rodent hippocampus more than half a century ago, and ever since have been considered as the biological basis of how the brain learns and remembers, and implicated in neurological problems such as addition (seriously, you can’t pass Neuro 101 without learning about LTP and LTD!).

So it’s perhaps especially salient that one of the first artificial-brain hybrid networks recapitulated this classic result.

To visualize: the three-neuron network began in Switzerland, with an artificial neuron with the badass name of “silicon spiking neuron.” That neuron is linked to an artificial synapse, a “memristor” located in the UK, which is then linked to a biological rat neuron cultured in Italy. The rat neuron has a “smart” microelectrode, controlled by the artificial synapse, to stimulate it. This is the artificial-to-biological pathway.

Meanwhile, the rat neuron in Italy also has electrodes that listen in on its electrical signaling. This signaling is passed back to another artificial synapse in the UK, which is then used to control a second artificial neuron back in Switzerland. This is the biological-to-artificial pathway back. As a testimony in how far we’ve come in digitizing neural signaling, all of the biological neural responses are digitized and sent over the internet to control its far-out artificial partner.

Here’s the crux: to demonstrate a functional neural network, just having the biological neuron passively “pass on” electrical stimulation isn’t enough. It has to show the capacity to learn, that is, to be able to mimic the amping up and down-tuning that are LTP and LTD, respectively.

You’ve probably guessed the results: certain stimulation patterns to the first artificial neuron in Switzerland changed how the artificial synapse in the UK operated. This, in turn, changed the stimulation to the biological neuron, so that it either amped up or toned down depending on the input.

Similarly, the response of the biological neuron altered the second artificial synapse, which then controlled the output of the second artificial neuron. Altogether, the biological and artificial components seamlessly linked up, over thousands of miles, into a functional neural circuit.

Cyborg Mind-Meld
So…I’m still picking my jaw up off the floor.

It’s utterly insane seeing a classic neuroscience learning experiment repeated with an integrated network with artificial components. That said, a three-neuron network is far from the thousands of synapses (if not more) needed to truly re-establish a broken neural circuit in the hippocampus, which DARPA has been aiming to do. And LTP/LTD has come under fire recently as the de facto brain mechanism for learning, though so far they remain cemented as neuroscience dogma.

However, this is one of the few studies where you see fields coming together. As Richard Feynman famously said, “What I cannot recreate, I cannot understand.” Even though neuromorphic chips were built on a high-level rather than molecular-level understanding of how neurons work, the study shows that artificial versions can still synapse with their biological counterparts. We’re not just on the right path towards understanding the brain, we’re recreating it, in hardware—if just a little.

While the study doesn’t have immediate use cases, practically it does boost both the neuromorphic computing and neuroprosthetic fields.

“We are very excited with this new development,” said study author Dr. Themis Prodromakis at the University of Southampton. “On one side it sets the basis for a novel scenario that was never encountered during natural evolution, where biological and artificial neurons are linked together and communicate across global networks; laying the foundations for the Internet of Neuro-electronics. On the other hand, it brings new prospects to neuroprosthetic technologies, paving the way towards research into replacing dysfunctional parts of the brain with AI chips.”

Image Credit: Gerd Altmann from Pixabay Continue reading →

Posted in Human Robots