Computers Math

Researchers at North Carolina State University have demonstrated a new method that leverages artificial intelligence (AI) and computer simulations to train robotic exoskeletons to autonomously help users save energy while walking, running and climbing stairs.
“This work proposes and demonstrates a new machine-learning framework that bridges the gap between simulation and reality to autonomously control wearable robots to improve mobility and health of humans,” says Hao Su, corresponding author of a paper on the work which will be published June 12 in the journal Nature.
“Exoskeletons have enormous potential to improve human locomotive performance,” says Su, who is an associate professor of mechanical and aerospace engineering at North Carolina State University. “However, their development and broad dissemination are limited by the requirement for lengthy human tests and handcrafted control laws.
“The key idea here is that the embodied AI in a portable exoskeleton is learning how to help people walk, run or climb in a computer simulation, without requiring any experiments,” says Su.
Specifically, the researchers focused on improving autonomous control of embodied AI systems — which are systems where an AI program is integrated into a physical robot technology. This work focused on teaching robotic exoskeletons how to assist able-bodied people with various movements. Normally, users have to spend hours “training” an exoskeleton so that the technology knows how much force is needed — and when to apply that force — to help users walk, run or climb stairs. The new method allows users to utilize the exoskeletons immediately.
“This work is essentially making science fiction reality — allowing people to burn less energy while conducting a variety of tasks,” says Su.
“We have developed a way to train and control wearable robots to directly benefit humans,” says Shuzhen Luo, first author of the paper and a former postdoctoral researcher at NC State. Luo is now an assistant professor at Embry-Riddle Aeronautical University.

For example, in testing with human subjects, the researchers found that study participants used 24.3% less metabolic energy when walking in the robotic exoskeleton than without the exoskeleton. Participants used 13.1% less energy when running in the exoskeleton, and 15.4% less energy when climbing stairs.
“It’s important to note that these energy reductions are comparing the performance of the robotic exoskeleton to that of a user who is not wearing an exoskeleton,” Su says. “That means it’s a true measure of how much energy the exoskeleton saves.”
While this study focused on the researchers’ work with able-bodied people, the new method also applies to robotic exoskeleton applications aimed at helping people with mobility impairments.
“Our framework may offer a generalizable and scalable strategy for the rapid development and widespread adoption of a variety of assistive robots for both able-bodied and mobility-impaired individuals,” Su says.
“We are in the early stages of testing the new method’s performance in robotic exoskeletons being used by older adults and people with neurological conditions, such as cerebral palsy. And we are also interested in exploring how the method could improve the performance of robotic prosthetic devices for amputee populations.”
This research was done with support from the National Science Foundation under awards 1944655 and 2026622; the National Institute on Disability, Independent Living, and Rehabilitation Research, under award 90DPGE0019 and Switzer Research Fellowship SFGE22000372; and the National Institutes of Health, under award 1R01EB035404.
Shuzhen Luo and Hao Su are co-inventors on intellectual property related to the controller discussed in this work. Su is also a co-founder of, and has a financial interest in, Picasso Intelligence, LLC, which develops exoskeletons. More

It is one of the most hotly debated topics in today’s workplace: Is allowing employees to log in from home a few days a week good for their productivity, careers, and job satisfaction?
Nicholas Bloom, a Stanford economist and one of the foremost researchers on work-from-home policies, has uncovered compelling evidence that hybrid schedules are a boon to both employees and their bosses.
In a study, newly published in the journal Nature, of an experiment on more than 1,600 workers at Trip.com — a Chinese company that is one of the world’s largest online travel agencies — Bloom finds that employees who work from home for two days a week are just as productive and as likely to be promoted as their fully office-based peers.
On a third key measure, employee turnover, the results were also encouraging. Resignations fell by 33 percent among workers who shifted from working full-time in the office to a hybrid schedule. Women, non-managers, and employees with long commutes were the least likely to quit their jobs when their treks to the office were cut to three days a week. Trip.com estimates that reduced attrition saved the company millions of dollars.
“The results are clear: Hybrid work is a win-win-win for employee productivity, performance, and retention,” says Bloom, who is the William D. Eberle Professor of Economics at the Stanford School of Humanities and Sciences and also a senior fellow at the Stanford Institute for Economic Policy Research (SIEPR).
The findings are especially significant given that, by Bloom’s count, about 100 million workers worldwide now spend a mix of days at home and in the office each week, more than four years after COVID-19 pandemic lockdowns upended how and where people do their jobs. Many of these hybrid workers are lawyers, accountants, marketers, software engineers and other with a college degree or higher.
Over time, though, working outside the office has come under attack from high-profile business leaders like Elon Musk, the head of Tesla, SpaceX, and X (formerly Twitter), and Jamie Dimon, CEO of JPMorgan Chase, who argue that the costs of remote work outweigh any benefits. Opponents say that employee training and mentoring, innovation, and company culture suffer when workers are not on site five days a week.

Blooms says that critics often confuse hybrid for fully remote, in part because most of the research into working from home has focused on workers who aren’t required to come into an office and on a specific type of job, like customer support or data entry. The results of these studies have been mixed, though they tend to skew negative. This suggests to Bloom that problems with fully remote work arise when it’s not managed well.
As one of the few randomized control trials to analyze hybrid arrangements — where workers are offsite two or three days a week and are in the office the rest of the time — Bloom says his findings offer important lessons for other multinationals, many of which share similarities with Trip.com.
“This study offers powerful evidence for why 80 percent of U.S. companies now offer some form of remote work,” Bloom says, “and for why the remaining 20 percent of firms that don’t are likely paying a price.”
The research is also the largest to date of hybrid work involving university-trained professionals that relies on the gold standard in research, the randomized controlled trial. This allowed Bloom and his co-authors to show that the benefits they identify resulted from Trip.com’s hybrid experiment and not something else.
In addition to Bloom, the study’s authors are Ruobing Han, an assistant professor at The Chinese University of Hong Kong, and James Liang, an economics professor at Peking University and co-founder of Trip.com. Han and Liang both earned their PhDs in economics from Stanford.
The hybrid approach: Only winners
Trip.com didn’t have a hybrid work policy when it undertook the 6-month experiment starting in 2021 that is at the heart of the study. In all, 395 managers and 1,217 non-managers with undergraduate degrees — all of whom worked in engineering, marketing, accounting, and finance in the company’s Shanghai office — participated. Employees whose birthdays fell on an even-numbered day of the month were told to come to the office five days a week. Workers with odd-numbered birthdays were allowed to work from home two days a week.

Of the study participants, 32 percent also had postgraduate degrees, mostly in computer science, accounting or finance. Most were in their mid-30s, half had children, and 65 percent were male.
In finding that hybrid work not only helps employees, but also companies, the researchers relied on various company data and worker surveys, including performance reviews and promotions for up to two years after the experiment. Trip.com’s thorough performance review process includes evaluations of an employee’s contributions to innovation, leadership, and mentoring.
The study authors also compared the quality and amount of computer code written by Trip.com software engineers who were hybrid against code produced by peers who were in the office full-time.
In finding that hybrid work had zero effect on workers’ productivity or career advancement and dramatically boosted retention rates, the study authors highlight some important nuances. Resignations, for example, fell only among non-managers; managers were just as likely to quit whether they were hybrid or not.
Bloom and his coauthors identify misconceptions held by workers and their bosses. Workers, especially women, were reluctant to sign up as volunteers for Trip.com’s hybrid trial — likely for fear that they would be judged negatively for not coming into the office five days a week, Bloom says. In addition, managers predicted on average that remote working would hurt productivity, only to change their minds by the time the experiment ended.
For business leaders, Bloom says the study confirms that concerns that hybrid work does more harm than good are overblown.
“If managed right, letting employees work from home two or three days a week still gets you the level of mentoring, culture-building, and innovation that you want,” Bloom says. “From an economic policymaking standpoint, hybrid work is one of the few instances where there aren’t major trade-offs with clear winners and clear losers. There are almost only winners.”
Trip.com was sold: It now allows hybrid work company-wide. More

An artificial intelligence-powered virtual teammate with a female voice boosts participation and productivity among women on teams dominated by men, according to new Cornell University research.
The findings suggest that the gender of an AI’s voice can positively tweak the dynamics of gender-imbalanced teams and could help inform the design of bots used for human-AI teamwork, researchers said.
The findings mirror previous research that shows minority teammates are more likely to participate if the team adds members similar to them, said Angel Hsing-Chi Hwang, postdoctoral associate in information science and lead author of the paper.
To better understand how AI can help gender-imbalanced teams, Hwang and Andrea Stevenson Won, associate professor of communication and the paper’s co-author, carried out an experiment with around 180 men and women who were assigned to groups of three and asked to collaborate virtually on a set of tasks (the study only included participants who identified as either male or female).
Each group had either one woman or one man and a fourth agent in the form of an abstract shape with either a male or female voice, which would appear on screen and read instructions, contribute an idea and handle timekeeping. There was a catch — the bot wasn’t completely automated. In what’s referred to in human-computer interaction as a “Wizard of Oz” experiment, Hwang was behind the scenes, feeding lines generated by ChatGPT into the bot.
After the experiment, Hwang and Won analyzed the chat logs of team conversations to determine how often participants offered ideas or arguments. They also asked participants to reflect on the experience.
“When we looked at participants’ actual behaviors, that’s where we started to see differences between men and women and how they were reacting when there was either a female agent or a male agent on the team,” she said.
“One interesting thing about this study is that most participants didn’t express a preference for a male- or female-sounding voice,” Won said. “This implies that people’s social inferences about AI can be influential even when people don’t believe they are important.”
When women were in the minority, they participated more when the AI’s voice was female, while men in the minority were more talkative but were less focused on tasks when working with a male-sounding bot, researchers found. Unlike the men, women reported significantly more positive perceptions of the AI teammate when women were the minority members, according to researchers.
“With only a gendered voice, the AI agent can provide a small degree of support to women minority members in a group,” said Hwang. More

Researchers from North Carolina State University have demonstrated miniature soft hydraulic actuators that can be used to control the deformation and motion of soft robots that are less than a millimeter thick. The researchers have also demonstrated that this technique works with shape memory materials, allowing users to repeatedly lock the soft robots into a desired shape and return to the original shape as needed.
“Soft robotics holds promise for many applications, but it is challenging to design the actuators that drive the motion of soft robots on a small scale,” says Jie Yin, corresponding author of a paper on the work and an associate professor of mechanical and aerospace engineering at NC State. “Our approach makes use of commercially available multi-material 3D printing technologies and shape memory polymers to create soft actuators on a microscale that allow us to control very small soft robots, which allows for exceptional control and delicacy.”
The new technique relies on creating soft robots that consist of two layers. The first layer is a flexible polymer that is created using 3D printing technologies and incorporates a pattern of microfluidic channels — essentially very small tubes running through the material. The second layer is a flexible shape memory polymer. Altogether, the soft robot is only 0.8 millimeters thick.
By pumping fluid into the microfluidic channels, users create hydraulic pressure that forces the soft robot to move and change shape. The pattern of microfluidic channels controls the motion and shape change of the soft robot — whether it bends, twists, or so on. In addition, the amount of fluid being introduced, and how quickly it is introduced, controls how quickly the soft robot moves and the amount of force the soft robot exerts.
If users wish to ‘freeze’ the soft robot’s shape, they can apply moderate heat (64C, or 147F), and then let the robot cool briefly. This prevents the soft robot from reverting to its original shape, even after the liquid in the microfluidic channels is pumped out. If users want to return the soft robot to its original shape, they simply apply the heat again after pumping out the liquid, and the robot relaxes to its original configuration.
“A key factor here is fine-tuning the thickness of the shape memory layer relative to the layer that contains the microfluidic channels,” says Yinding Chi, co-lead author of the paper and a former Ph.D. student at NC State. “You need the shape memory layer to be thin enough to bend when the actuator’s pressure is applied, but thick enough to get the soft robot to retain its shape even after the pressure is removed.”
To demonstrate the technique, the researchers created a soft robot “gripper,” capable of picking up small objects. The researchers applied hydraulic pressure, causing the gripper to pinch closed on an object. By applying heat, the researchers were able to fix the gripper in its “closed” position, even after releasing pressure from the hydraulic actuator. The gripper could then be moved — transporting the object it held — into a new position. Researchers then applied heat again, causing the gripper to release the object it had picked up. Video of these soft robots in action can be found at https://youtu.be/5SIwsw9IyIc.

“Because these soft robots are so thin, we can heat them up to 64C quickly and easily using a small infrared light source — and they also cool very quickly,” says Haitao Qing, co-lead author of the paper and a Ph.D. student at NC State. “So this entire series of operations only takes about two minutes.
“And the movement does not have to be a gripper that pinches,” says Qing. “We’ve also demonstrated a gripper that was inspired by vines in nature. These grippers quickly wrap around an object and clasp it tightly, allowing for a secure grip.
“This paper serves as a proof-of-concept for this new technique, and we’re excited about potential applications for this class of miniature soft actuators in small-scale soft robots, shape-shifting machines, and biomedical engineering.”
This work was done with support from the National Science Foundation under grants 2126072 and 2329674. More

Virtual reality technology can do more than teach weaponry skills in law enforcement and military personnel, a new study suggests: It can accurately record shooting performance and reliably track individuals’ progress over time.
In the study of 30 people with a range of experience levels in handling a rifle, researchers at The Ohio State University found that a ballistic simulator captured data on the shooters’ accuracy, decision-making and reaction time — down to the millimeter in distance and millisecond in time — on a consistent basis.
In addition to confirming that the simulator — called the VirTra V-100 — is a dependable research tool, the findings could lead to establishing the first-ever standardized performance scores for virtual reality ballistics training.
“To our knowledge, we’re the first team to answer the question of whether the simulator could be converted to an assessment tool and if it’s credible to use it day-to-day,” said Alex Buga, first author of the study and a PhD student in kinesiology at Ohio State.
“We’ve figured out how to export the data and interpret it. We’ve focused on the three big challenges of marksmanship, decision-making and reaction time to measure 21 relevant variables — allowing us to put a report in a user’s hand and say, ‘This is how accurate, precise, focused and fast you are.'”
The study was published online June 6 in The Journal of Strength and Conditioning Research.
U.S. military leaders and law enforcement agencies have shown an interest in increasing the use of virtual reality for performance assessment, said Buga and senior study author Jeff Volek, professor of human sciences at Ohio State. Earlier this year, an Ohio Attorney General Task Force on the Future of Police Training in Ohio recommended incorporating virtual reality technology into training protocols.

Volek is the principal investigator on a $10 million U.S. Department of Defense grant focused on improving the health of military service members, veterans and the American public. As part of that initiative, the research team is investigating the extent to which nutritional ketosis reduces detrimental effects of sleep loss on cognitive and physical performance in ROTC cadets — including their shooting ability as measured by the VirTra simulator. Verifying the simulator’s results for research purposes triggered the attempt to extract and analyze its data.
“We were using it as an outcome variable for research, and we found that it has very good day-to-day reproducibility of performance, which is crucial for research,” Volek said. “You want a sensitive and reproducible outcome in your test where there’s not a lot of device or equipment variation.”
Because the lab also focuses on human performance in first responders, researchers’ conversations with military and law enforcement communities convinced Buga that data collected by the simulator could be more broadly useful.
“I created a few programs that enabled us to calculate the shooting data and produce objective training measures,” he said. “This equipment is close to what the military and police use every day, so this has potential to be used as a screening tool across the country.”
Users of the simulator operate the infrared-guided M4 rifle by shooting at a large screen onto which different digitally generated visuals are projected — no headset required. The rifle at Ohio State has been retrofitted to produce the same recoil as a police or military weapon.
The study participants included civilians, police and SWAT officers, and ROTC cadets. Each was first familiarized in a single learning session with the simulator and then completed multiple rounds of three different tasks in each of three study performance sessions.

In the first task, participants fired at the same target a total of 50 times to produce measures of shooting precision. The decision-making assessment involved shooting twice within two seconds at designated shapes and colors on a screen displaying multiple shape and color choices. In the reaction-time scenario, participants shot at a series of plates from left to right as rapidly as possible.
Internal consistency ratings showed the simulator generated good to excellent test-retest agreement on the 21 variables measured.
All participants were well-rested and completed the study sessions at about the same time of day. Self-evaluations showed that participants’ overall confidence about their shooting performance increased from their first to final sessions. They also rated the simulator as a realistic and a low-stress shooting assessment tool.
The low stress and well-rested conditions were important to establishing baseline performance measures, the researchers noted, which then would enable evaluating how injuries and other physical demands of first-responder professions affect shooting performance.
“This simulator could be used to assess the effectiveness of specific training programs designed to improve shooting performance, or to evaluate marksmanship in response to various stressors encountered by the same law enforcement and military personnel,” Buga said. “These novel lines of evidence have enabled us to push the boundaries of tactical research and set the groundwork for using virtual reality in sophisticated training scenarios that support national defense goals.”
Additional co-authors, all from Ohio State, included Drew Decker, Bradley Robinson, Christopher Crabtree, Justen Stoner, Lucas Arce, Xavier El-Shazly, Madison Kackley, Teryn Sapper, John Paul Anders and William Kraemer. More

Today, researchers are developing ways to accelerate discovery by combining automated experiments, artificial intelligence and high-performance computing. A novel tool developed at Oak Ridge National Laboratory that leverages those technologies has demonstrated that AI can influence materials synthesis and conduct associated experiments without human supervision.
This autonomous materials synthesis tool uses pulsed laser deposition, or PLD, to deposit a thin layer of substance onto a base material. It then employs AI to analyze how the quality of the newly created material relates to the synthesis conditions, such as temperature, pressure and energy emitted during the PLD process. The AI suggests a revised set of conditions that may yield improved quality and then controls the PLD equipment to conduct the next experiment.
“We built computer control of all processes into the system and incorporated some hardware innovations to enable AI to drive experimentation,” said the study’s leader, Sumner Harris of the Center for Nanophase Materials Sciences at ORNL. “The automation allows us to perform our work 10 times faster, and the AI can understand huge parameter spaces with far fewer samples.” More

The need to transition away from fossil fuels to more sustainable energy production is critical. That’s why a team of Concordia researchers is looking at a potential power source that not only produces no carbon emissions but removes carbon as it works: algae.
Researchers at the Optical-Bio Microsystems Lab recently published a new paper on this topic in the journal Energies. In it, they describe their method of extracting energy from the photosynthesis process of algae suspended in a specialized solution and housed in small power cells. Configured correctly, these cells can generate enough energy to power low- and ultra-low power devices such as Internet of Things (IoT) sensors.
“The idea of the micro photosynthetic power cell is to extract electrons produced through the process of photosynthesis,” says Kirankumar Kuruvinashetti, PhD 20, now a Mitacs postdoctoral associate at the University of Calgary.
“Photosynthesis produces oxygen and electrons. Our model traps the electrons, which allows us to generate electricity. So more than being a zero-emission technology, it’s a negative carbon emission technology: it absorbs carbon dioxide from the atmosphere and gives you a current. Its only byproduct is water.”
Power generated day and night
The micro photosynthetic power cell consists of an anode and a cathode chamber separated by a honeycomb-shaped proton exchange membrane. The researchers fabricated microelectrodes on both sides of the membrane to collect the charges released by the algae during photosynthesis. Each chamber measures only two centimetres by two centimetres by four millimetres.
The algae are suspended in a two-millilitre solution in the anode chamber while the cathode is filled with potassium ferricyanide, a type of electron acceptor. Once the algae undergo photosynthesis and begin to release electrons, the electrons will be collected through the membrane’s electrodes and conducted, creating a current.

The protons, meanwhile, will pass through the membrane into the cathode and cause oxidation, resulting in a potassium ferrocyanide reduction.
The process also works without direct sunlight, though at a lower intensity, explains PhD candidate and paper co-author Dhilippan Panneerselvam.
“Just like humans, algae are constantly breathing — but they intake carbon dioxide and release oxygen. Due to their photosynthesis machinery, they also release electrons during respiration. The electricity generation is not stopped. The electrons are continuously harvested.”
Muthukumaran Packirisamy, professor in the Department of Mechanical, Industrial and Aerospace Engineering and the paper’s corresponding author, admits the system is not yet able to compete in power generation with others like photovoltaic cells. The maximum possible terminal voltage of a single micro photosynthetic power cell is only 1.0V.
But he believes that, with enough research and development, including artificial intelligence-assisted integration technologies, this technology has the potential to be a viable, affordable and clean power source in the future.
It also offers significant manufacturing advantages over other systems, he says.
“Our system does not use any of the hazardous gases or microfibres needed for the silicon fabrication technology that photovoltaic cells rely on. Furthermore, disposing of silicon computer chips is not easy. We use biocompatible polymers, so the whole system is easily decomposable and very cheap to manufacture.” More

The agility with which humans and animals move is an evolutionary marvel that no robot has yet been able to closely emulate. To help probe the mystery of how brains control movement, Harvard neuroscientists have created a virtual rat with an artificial brain that can move around just like a real rodent.
Bence Ölveczky, professor in the Department of Organismic and Evolutionary Biology, led a group of researchers who collaborated with scientists at Google’s DeepMind AI lab to build a biomechanically realistic digital model of a rat. Using high-resolution data recorded from real rats, they trained an artificial neural network — the virtual rat’s “brain” — to control the virtual body in a physics simulator called MuJoco, where gravity and other forces are present.
Publishing in Nature, the researchers found that activations in the virtual control network accurately predicted neural activity measured from the brains of real rats producing the same behaviors, said Ölveczky, who is an expert at training (real) rats to learn complex behaviors in order to study their neural circuitry. The feat represents a new approach to studying how the brain controls movement, Ölveczky said, by leveraging advances in deep reinforcement learning and AI, as well as 3D movement-tracking in freely behaving animals.
The collaboration was “fantastic,” Ölveczky said. “DeepMind had developed a pipeline to train biomechanical agents to move around complex environments. We simply didn’t have the resources to run simulations like those, to train these networks.”
Working with the Harvard researchers was, likewise, “a really exciting opportunity for us,” said co-author and Google DeepMind Senior Director of Research Matthew Botvinick. “We’ve learned a huge amount from the challenge of building embodied agents: AI systems that not only have to think intelligently, but also have to translate that thinking into physical action in a complex environment. It seemed plausible that taking this same approach in a neuroscience context might be useful for providing insights in both behavior and brain function.”
Graduate student Diego Aldarondo worked closely with DeepMind researchers to train the artificial neural network to implement what are called inverse dynamics models, which scientists believe our brains use to guide movement. When we reach for a cup of coffee, for example, our brain quickly calculates the trajectory our arm should follow and translates this into motor commands. Similarly, based on data from actual rats, the network was fed a reference trajectory of the desired movement and learned to produce the forces to generate it. This allowed the virtual rat to imitate a diverse range of behaviors, even ones it hadn’t been explicitly trained on.
These simulations may launch an untapped area of virtual neuroscience in which AI-simulated animals, trained to behave like real ones, provide convenient and fully transparent models for studying neural circuits, and even how such circuits are compromised in disease. While Ölveczky’s lab is interested in fundamental questions about how the brain works, the platform could be used, as one example, to engineer better robotic control systems.
A next step might be to give the virtual animal autonomy to solve tasks akin to those encountered by real rats. “From our experiments, we have a lot of ideas about how such tasks are solved, and how the learning algorithms that underlie the acquisition of skilled behaviors are implemented,” Ölveczky continued. “We want to start using the virtual rats to test these ideas and help advance our understanding of how real brains generate complex behavior.” More