AI & Robotics News and Discussions

[wjfox]

[Yuli Ban]

Unitree’s Go1 Robot Dog Looks Pretty Great, Costs Just USD $2700, compared to $72K for Boston Dynamics’ Spot

In 2017, we first wrote about the Chinese startup Unitree Robotics, which had the goal of “making legged robots as popular and affordable as smartphones and drones.” Relative to the cost of other quadrupedal robots (like Boston Dynamics’ $74,000 Spot), Unitree’s quadrupeds are very affordable, with their A1 costing under $10,000 when it became available in 2020. This hasn’t quite reached the point of consumer electronics that Unitree is aiming for, but they’ve just gotten a lot closer: now available is the Unitree Go1, a totally decent looking small size quadruped that can be yours for an astonishingly low $2700.

[Yuli Ban]

Google says its artificial intelligence is faster and better than humans at laying out chips for artificial intelligence

Google claims not only has it made an AI that's faster and as good as if not better than humans at designing chips, the web giant is using it to design chips for faster and better AI.

By designing, we mean the drawing up of a chip's floorplan, which is the arrangement of its subsystems – such as its CPU and GPU cores, cache memory, RAM controllers, and so on – on its silicon die. The placement of the minute electronic circuits that make up these modules can affect the microchip's power consumption and processing speed: the wiring and signal routing needed to connect it all up matters a lot.

In a paper to be published this week in Nature, and seen by The Register ahead of publication, Googlers Azalia Mirhoseini and Anna Goldie, and their colleagues, describe a deep reinforcement-learning system that can create floorplans in under six hours whereas it can take human engineers and their automated tools months to come up with an optimal layout.

[Time_Traveller]

Training robots to manipulate soft and deformable objects

June 10, 2021



Robots can solve a Rubik's cube and navigate the rugged terrain of Mars, but they struggle with simple tasks like rolling out a piece of dough or handling a pair of chopsticks. Even with mountains of data, clear instructions, and extensive training, they have a difficult time with tasks easily picked up by a child.

A new simulation environment, PlasticineLab, is designed to make robot learning more intuitive. By building knowledge of the physical world into the simulator, the researchers hope to make it easier to train robots to manipulate real-world objects and materials that often bend and deform without returning to their original shape. Developed by researchers at MIT, the MIT-IBM Watson AI Lab, and University of California at San Diego, the simulator was launched at the International Conference on Learning Representations in May.

In PlasticineLab, the robot agent learns how to complete a range of given tasks by manipulating various soft objects in simulation. In RollingPin, the goal is to flatten a piece of dough by pressing on it or rolling over it with a pin; in Rope, to wind a rope around a pillar; and in Chopsticks, to pick up a rope and move it to a target location.

The researchers trained their agent to complete these and other tasks faster than agents trained under reinforcement-learning algorithms, they say, by embedding physical knowledge of the world into the simulator, which allowed them to leverage gradient descent-based optimization techniques to find the best solution.

"Programming a basic knowledge of physics into the simulator makes the learning process more efficient," says the study's lead author, Zhiao Huang, a former MIT-IBM Watson AI Lab intern who is now a Ph.D. student at the University of California at San Diego. "This gives the robot a more intuitive sense of the real world, which is full of living things and deformable objects."

https://techxplore.com/news/2021-06-rob ... mable.html

[Time_Traveller]

New processes for automated fabrication of fiber and silicone composite structures for soft robotics

June 9, 2021

Researchers from the Singapore University of Technology and Design (SUTD) have developed novel techniques, known as Automated Fiber Embedding (AFE), to produce complex fiber and silicone composite structures for soft robotics applications. Their work was published in IEEE Robotics and Automation Letters.

Many soft robot components, including sensors and actuators, utilize embedded continuous fibers within elastomeric substrates to achieve various functionalities. However, manual embedding of continuous fibers in soft substrates is challenging due to the complexities involved in handling precise layering, and retaining of the fibers in the patterned positions which are prone to inconsistencies.

In contrast, the AFE approaches developed by the research team led by Assistant Professor Pablo Valdivia y Alvarado, enabled high precision fabrication of complex layered composites without manual user intervention, thus significantly augmenting the range of fabrication possibilities while saving time and labor.

The techniques exploited seamless combinations of fiber embedding with elastomeric deposition via Direct Ink Writing in an automated manner and allowed precise control of depth and fiber spacing within composite structures. This process automation has great potential for the fabrication and tailoring of soft robot components which require complex geometries that cannot be easily achieved manually.

https://techxplore.com/news/2021-06-aut ... osite.html

[Time_Traveller]

A social robot that could help children to regulate their emotions

June 9, 2021



In recent years, roboticists have developed a broad variety of social robots, robots designed to communicate with humans, assist them and support them in several different ways. This includes robotic toys and other robots designed to be used by children.

Researchers at University of California- Santa Cruz (UCSC), King's College London, and a US-based company called Sproutel, and Committee for Children have recently developed a new socially assistive robot specifically designed to aid emotional regulation in children. This robot, presented in a paper pre-published on arXiv, resembles a small creature that a child might want to care for or cuddle.

"My research team at University of California Santa Cruz had been working on designing smart fidget devices and understanding use of fidget objects for a while and I met Petr Slovak (a key collaborator on the work in the paper) at a conference workshop," Katherine Isbister, one of the researchers who carried out the study, told TechXplore. "His interests were focused on social emotional learning and how to scaffold those skills."

Slovak, a lecturer and researcher at King's College London, thought that smart fidget devices could aid emotional learning and emotional regulation in children. He thus started collaborating with Isbister, as well as other students and researchers at UCSC, to develop a robot or device that could be used by children to regulate their emotions.

The robot developed by Isbister, Slovak and their colleagues looks a bit like a plush toy or stuffed animal. This 'robotic stuffed toy' responds to a child's touch, calming down and reducing its movements as a child gently caresses it or hugs it.

https://techxplore.com/news/2021-06-soc ... tions.html

[funkervogt]

Chip floorplanning is the engineering task of designing the physical layout of a computer chip. Despite five decades of research1, chip floorplanning has defied automation, requiring months of intense effort by physical design engineers to produce manufacturable layouts. Here we present a deep reinforcement learning approach to chip floorplanning. In under six hours, our method automatically generates chip floorplans that are superior or comparable to those produced by humans in all key metrics, including power consumption, performance and chip area.

https://www.nature.com/articles/s41586-021-03544-w.epdf

Reminds me of this:

[Yuli Ban]

U.S. Launches Task Force to Study Opening Government Data for AI Research

The Biden administration launched an initiative Thursday aiming to make more government data available to artificial intelligence researchers, part of a broader push to keep the U.S. on the cutting edge of the crucial new technology.

The National Artificial Intelligence Research Resource Task Force, a group of 12 members from academia, government, and industry led by officials from the White House Office of Science and Technology Policy and the National Science Foundation, will draft a strategy for creating an AI research resource that could, in part, give researchers secure access to stores of anonymous data about Americans, from demographics to health and driving habits.

They would also look to make available computing power to analyze the data, with the goal of allowing access to researchers across the country.

“This is a moment that is calling us to be strengthening our speed and scale” when it comes to advances in AI technology, said National Science Foundation Director Sethuraman Panchanathan in an interview. “It is also calling us to make sure that innovation is everywhere.”

Starspawn0's response to whether this would make a real difference:

Probably not much. What would have a HUGE economic impact would be if they poured $1 billion into constructing large robotics datasets -- say, hire teams of humans to remotely control a large number of robots in a large number of settings and tasks. We don't have that kind of data. If we had it, robotics would go through several quantum leaps in no time.

To be on the safe side (i.e. that they don't want the money to go to waste), they could do this in an incremental fashion -- say, begin by spending $50 million, and see how good the robots that result. Then, bump it up to another $100 million if the first round was successful. And if successful, again, keep bumping it up.

[weatheriscool]

Google Deploys AI to Build Better AI Hardware Accelerators
https://www.extremetech.com/extreme/323 ... celerators
By Joel Hruska on June 11, 2021 at 7:00 am

Google reports that it is now using AI to build its future Tensor Processing Units. The company has published some work in this area about a year ago, but today’s announcement indicates the technology has matured. Alexis Mirhoseini led the project.

The semiconductor industry has invested in various tools that automate parts of the design process for decades. Back when a CPU had 10,000 to 100,000 transistors, hand-drawn floor plans and circuit layouts were the only way to build a chip. Today, much of the design work is automated, though engineers may still be used in specific, critical paths.

Google is claiming it can adopt AI to help with floorplanning. The floorplan of a microprocessor — literally, its physical layout — has historically been a difficult task to automate. Even with the aid of modern software tools, laying out a new floorplan can take weeks. Over many decades, a great deal of work has gone into building software to better handle this complex problem, but humans are still integral to the process. Now, Google is claiming its new AI can do the job in a matter of hours.

[weatheriscool]

Study explores the potential of using a humanoid robot to entertain the elderly
https://techxplore.com/news/2021-06-exp ... derly.html
by Ingrid Fadelli , Tech Xplore

Humanoid robots have the potential of assisting humans in a variety of settings, ranging from home environments to malls, schools and healthcare facilities. Some roboticists have been specifically investigating the potential of social robots as tools to offer care and companionship to the elderly population.

Researchers at Nayang Technological University have recently carried out a study exploring the potential of a humanoid robot for entertaining residents of an elderly care home. Their paper, pre-published on arXiv, specifically examined the reactions of a group of elderly individuals as they played Bingo with a social robot called Nadine.

"The main goal of our paper was to investigate whether a robot with human appearance and gestures can support the elderly, particularly by entertaining them with games, such as Bingo," Nadia Magnenat Thalmann, one of the researchers who carried out the study, told TechXplore. "We wanted to find out whether these kinds of robots can help to decrease loneliness among the elderly, offering a presence and stimulus by playing games with them at any time of the day."

[Yuli Ban]

It’s (Still) Really Hard for Robots to Autonomously Do Household Chores

Every time we think that we’re getting a little bit closer to a household robot, new research comes out showing just how far we have to go. Certainly, we’ve seen lots of progress in specific areas like grasping and semantic understanding and whatnot, but putting it all together into a hardware platform that can actually get stuff done autonomously still seems quite a way off.

In a paper presented at ICRA 2021 this month, researchers from the University of Bremen conducted a “Robot Household Marathon Experiment,” where a PR2 robot was tasked with first setting a table for a simple breakfast and then cleaning up afterwards in order to “investigate and evaluate the scalability and the robustness aspects of mobile manipulation.” While this sort of thing kinda seems like something robots should have figured out, it may not surprise you to learn that it’s actually still a significant challenge.

PR2’s job here is to prepare breakfast by bringing a bowl, a spoon, a cup, a milk box, and a box of cereal to a dining table. After breakfast, the PR2 then has to place washable objects into the dishwasher, put the cereal box back into its storage location, toss the milk box into the trash. The objects vary in shape and appearance, and the robot is only given symbolic descriptions of object locations (in the fridge, on the counter). It’s a very realistic but also very challenging scenario, which probably explains why it takes the poor PR2 90 minutes to complete it.

Because it's still really hard for AI to do these things too. Current systems like GPT-3 show signs of generality in limited areas, but we currently have no general world-model that can do multiple tasks. We still have to daisy chain narrow systems together to get a robot to function in the real world, and each of those narrow systems can easily break. They don't feed back into each other like biological general intelligence can, and aren't very adaptable either.

And then you have power too! Batteries are better than they were even ten years ago, but we still need plenty of improvements to make domestic robots that can work quickly enough for long enough to act as autonomous servants.

[Yuli Ban]

Experts Doubt Ethical AI Design Will Be Broadly Adopted As The Norm Within The Next Decade
A Majority Worries That The Evolution Of Artificial Intelligence By 2030 Will Continue To Be Primarily Focused On Optimizing Profits And Social Control

Artificial intelligence systems “understand” and shape a lot of what happens in people’s lives. AI applications “speak” to people and answer questions when the name of a digital voice assistant is called out. They run the chatbots that handle customer-service issues people have with companies. They help diagnose cancer and other medical conditions. They scour the use of credit cards for signs of fraud, and they determine who could be a credit risk.

They help people drive from point A to point B and update traffic information to shorten travel times. They are the operating system of driverless vehicles. They sift applications to make recommendations about job candidates. They determine the material that is offered up in people’s newsfeeds and video choices.

They recognize people’s faces, translate languages and suggest how to complete people’s sentences or search queries. They can “read” people’s emotions. They beat them at sophisticated games. They write news stories, paint in the style of Vincent Van Gogh and create music that sounds quite like the Beatles and Bach.

Corporations and governments are charging evermore expansively into AI development. Increasingly, nonprogrammers can set up off-the-shelf, pre-built AI tools as they prefer.

As this unfolds, a number of experts and advocates around the world have become worried about the long-term impact and implications of AI applications.

[Yuli Ban]

Spot the robot dog's owner, Boston Dynamics, officially sold to Hyundai

Spot the friendly robot dog has a new owner. Well, the company that makes this computer-controlled pooch does. On Monday, SoftBank announced Hyundai Motor Group officially took a controlling stake in Boston Dynamics, which tinkers with robots like Spot. Hyundai and SoftBank first revealed the deal last year, but as of today, Hyundai now owns a controlling stake -- at least 80% of the firm.

The companies didn't provide any additional details surrounding the deal, but we know Hyundai plopped down $1.1 billion to make this sale happen. As for what Boston Dynamics' new owner wants from the company, that's also totally unclear. The best we can assume is that Hyundai's pretty interested in robots. The automaker previously announced a new division dedicated to creating "walking cars" and other robots, so it could be the company's keen to have Boston Dynamics on its team as it pursues that. I can see it now: "Every new Hyundai walking car comes with your very own robot dog!"

That's probably not the case, but still, it's a massive acquisition for the Korean automaker and we're mighty curious to see what Hyundai plans to do with this robotics firm. If anything, the company could help Hyundai design new robots for manufacturing and logistics. That's not as fun, but could be highly valuable nonetheless.

[weatheriscool]

Intelligent carpet gives insight into human poses
https://techxplore.com/news/2021-06-int ... poses.html
by Rachel Gordon, Massachusetts Institute of Technology

The sentient magic carpet from 'Aladdin' might have a new competitor. While it can't fly or speak, a new tactile sensing carpet from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) can estimate human poses without using cameras, in a step towards improving self-powered personalized healthcare, smart homes, and gaming.

Many of our daily activities involve physical contact with the ground: walking, exercising, or resting. These embedded interactions contain a wealth of information that help us better understand people's movements.

Previous research has leveraged use of single RGB cameras, (think Microsoft Kinect), wearable omnidirectional cameras, and even plain old off the shelf webcams, but with the inevitable byproducts of camera occlusions and privacy concerns.

[weatheriscool]

Toyota announces new capabilities for domestic robots

by Bob Yirka , Tech Xplore
https://techxplore.com/news/2021-06-toy ... obots.html

Toyota Research Institute (TRI) has announced on its press blog that engineers with the company have developed new capabilities for its line of domestic robots. The new capabilities include recognizing and manipulating transparent objects and wiping down counters and tables. In the video accompanying the announcement, a robot grabs a smartphone off a table and uses it to take a selfie video as it carries on with cleaning a kitchen.

Toyota and other robotics companies in Japan have made clear their plans to create robots that will serve as domestic helpers in the coming years. Japan, notably, has one of the oldest populations and expects that there will be as many people over 65, as under by the year 2040. To address the aging problem, the robotics companies have been working to build robots that will be able to do the work that people find more challenging as they get older. To that end, engineers in Japan and other countries have designed robots that can sweep floors, vacuum carpets and even fold clothes. In this new effort, the team at TRI has tackled one of the difficulties that robot engineers have been facing—how to recognize and deal with transparent or shiny objects.

[Yuli Ban]

Google Trains Two Billion Parameter AI Vision Model
The model was trained on three billion images and achieved 90.45% top-1 accuracy on ImageNet, setting a new state-of-the-art record

Researchers at Google Brain announced a deep-learning computer vision (CV) model containing two billion parameters. The model was trained on three billion images and achieved 90.45% top-1 accuracy on ImageNet, setting a new state-of-the-art record.

The team described the model and experiments in a paper published on arXiv. The model, dubbed ViT-G/14, is based on Google's recent work on Vision Transformers (ViT). ViT-G/14 outperformed previous state-of-the-art solutions on several benchmarks, including ImageNet, ImageNet-v2, and VTAB-1k. On the few-shot image recognition task, the accuracy improvement was more than five percentage-points. The researchers also trained several smaller versions of the model to investigate a scaling law for the architecture, noting that the performance follows a power-law function, similar to Transformer models used for natural language processing (NLP) tasks.

[Yuli Ban]

The Pentagon Just Launched a New AI Initiative to Transform Global Warfare
In other words, the AI arms race is finally here

Imagine a nuclear arms race with few-to-no casualties and almost no threat of human extinction. Now replace nukes with unconscionable volumes of real-time data. Welcome to the future of global warfare.

The Pentagon just launched a new artificial intelligence (AI) initiative designed to enhance the curation process of massive amounts of tactical data, according to an announcement from Deputy Secretary of Defense Kathleen Hicks, in an initial report from National Defense.

In other words, the AI arms race is finally here.

AI will streamline the Pentagon's global command infrastructures
The new campaign, which is called the DoD AI and Data Acceleration initiative (ADA), is designed to move novel data and AI-linked concepts like joint all-domain command and control, also called JAD2, said Hicks in the report. "The ADA initiative will generate foundational capabilities through a series of implementation experiments or exercises, each one purposely building understanding through successive and incremental learning," she explained during the AI Symposium.

[Yuli Ban]

[Yuli Ban]

[weatheriscool]

Samsung's latest robot vac uses AI to recognize small obstacles
By Paul Ridden
June 25, 2021

https://newatlas.com/around-the-home/sa ... ot-vacuum/

Samsung is claiming a few world firsts with its latest robot vacuum cleaner, which was originally launched at CES 2021 back in January. The Jet Bot AI+ is said to be the first to feature an active stereo-type 3D sensor for object detection, and the first equipped with Intel AI for object recognition.

"We’re truly excited to introduce a robot vacuum that comes with industry-leading object detection and recognition technologies," said Samsung's Hyesoon Yang. "Consumers can now clean their homes more intelligently and efficiently using the AI-powered Jet Bot that makes cleaning more personalized and convenient."

The company's design team wanted to equip the automated cleaning assistant with enough smarts that it could "actively and efficiently clean like humans" and do so without much intervention from users.