Tag Archives: newest
#437820 In-Shoe Sensors and Mobile Robots Keep ...
In shoe sensor
Researchers at Stevens Institute of Technology are leveraging some of the newest mechanical and robotic technologies to help some of our oldest populations stay healthy, active, and independent.
Yi Guo, professor of electrical and computer engineering and director of the Robotics and Automation Laboratory, and Damiano Zanotto, assistant professor of mechanical engineering, and director of the Wearable Robotic Systems Laboratory, are collaborating with Ashley Lytle, assistant professor in Stevens’ College of Arts and Letters, and Ashwini K. Rao of Columbia University Medical Center, to combine an assistive mobile robot companion with wearable in-shoe sensors in a system designed to help elderly individuals maintain the balance and motion they need to thrive.
“Balance and motion can be significant issues for this population, and if elderly people fall and experience an injury, they are less likely to stay fit and exercise,” Guo said. “As a consequence, their level of fitness and performance decreases. Our mobile robot companion can help decrease the chances of falling and contribute to a healthy lifestyle by keeping their walking function at a good level.”
The mobile robots are designed to lead walking sessions and using the in-shoe sensors, monitor the user’s gait, indicate issues, and adjust the exercise speed and pace. The initiative is part of a four-year National Science Foundation research project.
“For the first time, we’re integrating our wearable sensing technology with an autonomous mobile robot,” said Zanotto, who worked with elderly people at Columbia University Medical Center for three years before coming to Stevens in 2016. “It’s exciting to be combining these different areas of expertise to leverage the strong points of wearable sensing technology, such as accurately capturing human movement, with the advantages of mobile robotics, such as much larger computational powers.”
The team is developing algorithms that fuse real-time data from smart, unobtrusive, in-shoe sensors and advanced on-board sensors to inform the robot’s navigation protocols and control the way the robot interacts with elderly individuals. It’s a promising way to assist seniors in safely doing walking exercises and maintaining their quality of life.
Bringing the benefits of the lab to life
Guo and Zanotto are working with Lytle, an expert in social and health psychology, to implement a social connectivity capability and make the bi-directional interaction between human and robot even more intuitive, engaging, and meaningful for seniors.
“Especially during COVID, it’s important for elderly people living on their own to connect socially with family and friends,” Zanotto said, “and the robot companion will also offer teleconferencing tools to provide that interaction in an intuitive and transparent way.”
“We want to use the robot for social connectedness, perhaps integrating it with a conversation agent such as Alexa,” Guo added. “The goal is to make it a companion robot that can sense, for example, that you are cooking, or you’re in the living room, and help with things you would do there.”
It’s a powerful example of how abstract concepts can have meaningful real-life benefits.
“As engineers, we tend to work in the lab, trying to optimize our algorithms and devices and technologies,” Zanotto noted, “but at the end of the day, what we do has limited value unless it has impact on real life. It’s fascinating to see how the devices and technologies we’re developing in the lab can be applied to make a difference for real people.”
Maintaining balance in a global pandemic
Although COVID-19 has delayed the planned testing at a senior center in New York City, it has not stopped the team’s progress.
“Although we can’t test on elderly populations yet, our students are still testing in the lab,” Guo said. “This summer and fall, for the first time, the students validated the system’s real-time ability to monitor and assess the dynamic margin of stability during walking—in other words, to evaluate whether the person following the robot is walking normally or has a risk of falling. They’re also designing parameters for the robot to give early warnings and feedback that help the human subjects correct posture and gait issues while walking.”
Those warnings would be literally underfoot, as the in-shoe sensors would pulse like a vibrating cell phone to deliver immediate directional information to the subject.
“We’re not the first to use this vibrotactile stimuli technology, but this application is new,” Zanotto said.
So far, the team has published papers in top robotics publication venues including IEEE Transactions on Neural Systems and Rehabilitation Engineering and the 2020 IEEE International Conference on Robotics and Automation (ICRA). It’s a big step toward realizing the synergies of bringing the technical expertise of engineers to bear on the clinical focus on biometrics—and the real lives of seniors everywhere. Continue reading
#437701 Robotics, AI, and Cloud Computing ...
IBM must be brimming with confidence about its new automated system for performing chemical synthesis because Big Blue just had twenty or so journalists demo the complex technology live in a virtual room.
IBM even had one of the journalists choose the molecule for the demo: a molecule in a potential Covid-19 treatment. And then we watched as the system synthesized and tested the molecule and provided its analysis in a PDF document that we all saw in the other journalist’s computer. It all worked; again, that’s confidence.
The complex system is based upon technology IBM started developing three years ago that uses artificial intelligence (AI) to predict chemical reactions. In August 2018, IBM made this service available via the Cloud and dubbed it RXN for Chemistry.
Now, the company has added a new wrinkle to its Cloud-based AI: robotics. This new and improved system is no longer named simply RXN for Chemistry, but RoboRXN for Chemistry.
All of the journalists assembled for this live demo of RoboRXN could watch as the robotic system executed various steps, such as moving the reactor to a small reagent and then moving the solvent to a small reagent. The robotic system carried out the entire set of procedures—completing the synthesis and analysis of the molecule—in eight steps.
Image: IBM Research
IBM RXN helps predict chemical reaction outcomes or design retrosynthesis in seconds.
In regular practice, a user will be able to suggest a combination of molecules they would like to test. The AI will pick up the order and task a robotic system to run the reactions necessary to produce and test the molecule. Users will be provided analyses of how well their molecules performed.
Back in March of this year, Silicon Valley-based startup Strateos demonstrated something similar that they had developed. That system also employed a robotic system to help researchers working from the Cloud create new chemical compounds. However, what distinguishes IBM’s system is its incorporation of a third element: the AI.
The backbone of IBM’s AI model is a machine learning translation method that treats chemistry like language translation. It translates the language of chemistry by converting reactants and reagents to products through the use of Statistical Machine Intelligence and Learning Engine (SMILE) representation to describe chemical entities.
IBM has also leveraged an automatic data driven strategy to ensure the quality of its data. Researchers there used millions of chemical reactions to teach the AI system chemistry, but contained within that data set were errors. So, how did IBM clean this so-called noisy data to eliminate the potential for bad models?
According to Alessandra Toniato, a researcher at IBM Zurichh, the team implemented what they dubbed the “forgetting experiment.”
Toniato explains that, in this approach, they asked the AI model how sure it was that the chemical examples it was given were examples of correct chemistry. When faced with this choice, the AI identified chemistry that it had “never learnt,” “forgotten six times,” or “never forgotten.” Those that were “never forgotten” were examples that were clean, and in this way they were able to clean the data that AI had been presented.
While the AI has always been part of the RXN for Chemistry, the robotics is the newest element. The main benefit that turning over the carrying out of the reactions to a robotic system is expected to yield is to free up chemists from doing the often tedious process of having to design a synthesis from scratch, says Matteo Manica, a research staff member in Cognitive Health Care and Life Sciences at IBM Research Zürich.
“In this demo, you could see how the system is synergistic between a human and AI,” said Manica. “Combine that with the fact that we can run all these processes with a robotic system 24/7 from anywhere in the world, and you can see how it will really help up to speed up the whole process.”
There appear to be two business models that IBM is pursuing with its latest technology. One is to deploy the entire system on the premises of a company. The other is to offer licenses to private Cloud installations.
Photo: Michael Buholzer
Teodoro Laino of IBM Research Europe.
“From a business perspective you can think of having a system like we demonstrated being replicated on the premise within companies or research groups that would like to have the technology available at their disposal,” says Teodoro Laino, distinguished RSM, manager at IBM Research Europe. “On the other hand, we are also pushing at bringing the entire system to a service level.”
Just as IBM is brimming with confidence about its new technology, the company also has grand aspirations for it.
Laino adds: “Our aim is to provide chemical services across the world, a sort of Amazon of chemistry, where instead of looking for chemistry already in stock, you are asking for chemistry on demand.”
< Back to IEEE COVID-19 Resources Continue reading
#437269 DeepMind’s Newest AI Programs Itself ...
When Deep Blue defeated world chess champion Garry Kasparov in 1997, it may have seemed artificial intelligence had finally arrived. A computer had just taken down one of the top chess players of all time. But it wasn’t to be.
Though Deep Blue was meticulously programmed top-to-bottom to play chess, the approach was too labor-intensive, too dependent on clear rules and bounded possibilities to succeed at more complex games, let alone in the real world. The next revolution would take a decade and a half, when vastly more computing power and data revived machine learning, an old idea in artificial intelligence just waiting for the world to catch up.
Today, machine learning dominates, mostly by way of a family of algorithms called deep learning, while symbolic AI, the dominant approach in Deep Blue’s day, has faded into the background.
Key to deep learning’s success is the fact the algorithms basically write themselves. Given some high-level programming and a dataset, they learn from experience. No engineer anticipates every possibility in code. The algorithms just figure it.
Now, Alphabet’s DeepMind is taking this automation further by developing deep learning algorithms that can handle programming tasks which have been, to date, the sole domain of the world’s top computer scientists (and take them years to write).
In a paper recently published on the pre-print server arXiv, a database for research papers that haven’t been peer reviewed yet, the DeepMind team described a new deep reinforcement learning algorithm that was able to discover its own value function—a critical programming rule in deep reinforcement learning—from scratch.
Surprisingly, the algorithm was also effective beyond the simple environments it trained in, going on to play Atari games—a different, more complicated task—at a level that was, at times, competitive with human-designed algorithms and achieving superhuman levels of play in 14 games.
DeepMind says the approach could accelerate the development of reinforcement learning algorithms and even lead to a shift in focus, where instead of spending years writing the algorithms themselves, researchers work to perfect the environments in which they train.
Pavlov’s Digital Dog
First, a little background.
Three main deep learning approaches are supervised, unsupervised, and reinforcement learning.
The first two consume huge amounts of data (like images or articles), look for patterns in the data, and use those patterns to inform actions (like identifying an image of a cat). To us, this is a pretty alien way to learn about the world. Not only would it be mind-numbingly dull to review millions of cat images, it’d take us years or more to do what these programs do in hours or days. And of course, we can learn what a cat looks like from just a few examples. So why bother?
While supervised and unsupervised deep learning emphasize the machine in machine learning, reinforcement learning is a bit more biological. It actually is the way we learn. Confronted with several possible actions, we predict which will be most rewarding based on experience—weighing the pleasure of eating a chocolate chip cookie against avoiding a cavity and trip to the dentist.
In deep reinforcement learning, algorithms go through a similar process as they take action. In the Atari game Breakout, for instance, a player guides a paddle to bounce a ball at a ceiling of bricks, trying to break as many as possible. When playing Breakout, should an algorithm move the paddle left or right? To decide, it runs a projection—this is the value function—of which direction will maximize the total points, or rewards, it can earn.
Move by move, game by game, an algorithm combines experience and value function to learn which actions bring greater rewards and improves its play, until eventually, it becomes an uncanny Breakout player.
Learning to Learn (Very Meta)
So, a key to deep reinforcement learning is developing a good value function. And that’s difficult. According to the DeepMind team, it takes years of manual research to write the rules guiding algorithmic actions—which is why automating the process is so alluring. Their new Learned Policy Gradient (LPG) algorithm makes solid progress in that direction.
LPG trained in a number of toy environments. Most of these were “gridworlds”—literally two-dimensional grids with objects in some squares. The AI moves square to square and earns points or punishments as it encounters objects. The grids vary in size, and the distribution of objects is either set or random. The training environments offer opportunities to learn fundamental lessons for reinforcement learning algorithms.
Only in LPG’s case, it had no value function to guide that learning.
Instead, LPG has what DeepMind calls a “meta-learner.” You might think of this as an algorithm within an algorithm that, by interacting with its environment, discovers both “what to predict,” thereby forming its version of a value function, and “how to learn from it,” applying its newly discovered value function to each decision it makes in the future.
Prior work in the area has had some success, but according to DeepMind, LPG is the first algorithm to discover reinforcement learning rules from scratch and to generalize beyond training. The latter was particularly surprising because Atari games are so different from the simple worlds LPG trained in—that is, it had never seen anything like an Atari game.
Time to Hand Over the Reins? Not Just Yet
LPG is still behind advanced human-designed algorithms, the researchers said. But it outperformed a human-designed benchmark in training and even some Atari games, which suggests it isn’t strictly worse, just that it specializes in some environments.
This is where there’s room for improvement and more research.
The more environments LPG saw, the more it could successfully generalize. Intriguingly, the researchers speculate that with enough well-designed training environments, the approach might yield a general-purpose reinforcement learning algorithm.
At the least, though, they say further automation of algorithm discovery—that is, algorithms learning to learn—will accelerate the field. In the near term, it can help researchers more quickly develop hand-designed algorithms. Further out, as self-discovered algorithms like LPG improve, engineers may shift from manually developing the algorithms themselves to building the environments where they learn.
Deep learning long ago left Deep Blue in the dust at games. Perhaps algorithms learning to learn will be a winning strategy in the real world too.
Image credit: Mike Szczepanski / Unsplash Continue reading