The lab turning commercial robots into rehab therapists

Prof. Shelly Levy-Tzedek  credit: Dani Machlis - Ben Gurion University of the Negev
Prof. Shelly Levy-Tzedek credit: Dani Machlis - Ben Gurion University of the Negev

Prof. Shelly Levy-Tzedek of Ben-Gurion University talks to "Globes" about human-machine relationships, the limits of AI, and why patients steer clear of robots that tell jokes.

As the health system becomes overburdened, budgets are denied, and employees leave in droves, there is a growing need to develop solutions that will enable the expansion of human capabilities - without humans being involved.

Consider, for example, a person undergoing surgery. The surgeon will operate for a few hours, but, later on, the patient may many need hours of physiotherapy to achieve the desired results. Because of insufficient staff and funding for physiotherapy, patients are sent home to exercise on their own. There they run into stumbling blocks, from becoming bored to carrying out exercises incorrectly in the absence of professional guidance.

Prof. Shelly Levy-Tzedek believes that at-home robots can take on some professional tasks, and is gradually proving this in her laboratory at Ben-Gurion University.

Levy-Tzedek, a biomedical engineer by training, with a PhD from MIT, is focused on developing social robots to aid in effective self-training. She uses commercially available robots, and programs them according to her needs. The term "social robots" refers to robots that can converse, demonstrate actions, remind people of tasks, assist in app use, and more. Social robots are gradually becoming more and more popular in the public space. "We worked with Pepper, for example, a humanoid robot that you sometimes see in hotels in Asia and at airports. It's used mainly for marketing purposes."

Levy-Tzedek programmed Pepper to help post-stroke patients, where physical therapy can go like something like this: the patient reaches for a cup, sensors signal to the robot the location of the cup, i.e. the degree of success. The robot can then give the patient instructions on what to do next: arrange the cups in order, switch them around, etc., at increasing levels of difficulty. There are other physical therapy modules, like hanging keys on a rack, and a memory exercise for recalling the names of objects.

"It should feel like playing a game, to increase motivation. It should be fun. Boredom, frustration and lack of feedback, are major factors in why people fail at doing physical therapy at home. One of our tasks is designed like an escape room. You have to find an ID card, type four digits, open a drawer, and find keys. In fact, you're performing a series of motor and cognitive functions from everyday life." Another task is designed like a game of poker or blackjack.

Not every robot needs artificial intelligence

In one experiment, Levy-Tzedek examined how the relationship with the robot affects long-term rehabilitative care outcomes. "Usually, academic experiments on robots test one-time encounters, but the advantage of robots is their ability to encourage perseverance," she says.

In the experiment, the subjects were divided into three groups: one exercised with a physiotherapist and received recommendations for exercise, but without a robot; the second received the same recommendations with the option to exercise with a robot; And the third received recommendations with support in the form of instructions on a computer screen.

"We observed that despite physiotherapy - for hand functionality - the control group did not improve at all. The group that received instructions via computer screen improved by 63%, and the robot group improved by 90%." This experiment was done as part of Dr. Ronit Feingold-Polak's doctoral thesis, advised by Levy-Tzedek.

Somewhat surprisingly, Levy-Tzedek's robots do not have a high levels of artificial intelligence (AI). "In the future, robots are expected to have limited levels of artificial intelligence capabilities; they will be able to detect and respond to compensatory movement patterns," she says. Something that is lacking in self-training lacks is feedback on these compensatory movements that cause patients to make incorrect movements in a drill on reaching for an object, for example. The lab is now writing an algorithm that will allow the robot to detect compensatory movements, offer suggestions, and also say "Well done" if a compensatory movement is not made.

If the robot isn't AI-enabled, does that ruin the illusion of it being intelligent?

"Not necessarily. We've had patients who actually treated it like a human being. For example, in our initial trails, it made mistakes sometimes, and there were people who said, 'I'm human and I make mistakes so, okay, it can make mistakes, too.' Others said they liked it and enjoyed meeting it. People said, 'He isn't judgmental', even though the physiotherapists weren't judgmental either. Maybe seeing themselves as others see them made them criticize themselves, and maybe when the robot starts giving feedback, it will be perceived as judgmental. That's something we'll have to check."

Judgment can also be positive. If you've made an effort, you may want an intelligent being to see and appreciate it.

"We record every event, and pass the information along to the physiotherapist so they can give the patient a real compliment for persistence."

A dedicated room has been set up for the system (robot and peripheral equipment) at the ADI Negev-Nahalat Eran Rehabilitation Village near Ofakim. Levy-Tzedek's laboratory is now working on a smaller, simpler, and cheaper home-use version. "My vision is that the product will also be available for rent from Yad Sarah [the national non-profit that loans medical and rehabilitative equipment and other services at nominal or no cost].

Adults feel connected - kids, not so much

Levy-Tzedek's lab studies human-robot interaction to learn how to construct them. "It's very important that the robot should not repeat phrases too often," says Levy-Tzedek, "but randomness isn't ideal either. For example, a joke shouldn't be told several times over. It feels annoying or in poor taste, and reinforces the fact that the robot isn't human. So, maybe partial randomness, and humor very rarely.

"We also observed that body movements had to be adapted to the local culture. For example, a robot originally from Japan made a motion that was supposed to indicate joy, and some Israelis asked, 'Why is he angry? What did I do wrong?'"

Levy-Tzedek also points to intergenerational differences. "Actually, the adults accepted it as an independent entity and said that they felt connected with it. Some of the young people did, but others said it was 'creepy'. We don't know exactly why. After all, these young people have grown up with technology and with people they know only via a screen, and yet they were the ones who said it was too weird for them. When we asked how they would like to tell the robot they had finished a training session - by pushing a button or stroking its hand - the adults preferred hand-stroking, and the younger ones, button-pushing."

A furry seal for pain reduction

In another experiment, Levy-Tzedek used a robot that was sold as a toy and shaped like a furry stuffed seal. "We examined healthy volunteers for pain endurance, with or without the robot present. We saw that they felt less pain and coped better when the robot was present and talking to them, and the situation improved even more when they petted or hugged it."

What makes some people feel like they're making a real connection with a robot? Is it related to that person's personality?

"We found no differences in the metrics we're familiar with. We assume that specific metrics can be developed for a willingness to accept the illusion of humanity in inanimate objects."

The robotic seal was also not AI-enabled. "It's very complicated to develop reliable artificial intelligence," admits Levy-Tzedek. "On the other hand, the illusion of attachment, as we have demonstrated, doesn't necessarily depend on it, and can maybe even destroy it. AI speech recognition for Hebrew is really bad. The ability to understand spoken Hebrew is almost non-existent, certainly when it comes to understanding different accents, or speech affected by stroke. The day when a robot will understand our patients, and they'll be able to hold a conversation with it, is still very far off."

In the near-term, even robots with English-language understanding won't convince anyone to suspend disbelief. Levy-Tzedek says. "I don't know of any robot that can really hold a conversation that could fool us." Therefore, she says, managing expectations is important. "In the end, the patient knows it's a training tool and not a friend, so there's no disappointment here. We don't expect to have a real conversation with our medical device, although the fact that it is humanoid is part of what does the job. This duality is something we don't fully understand."

Which is better - VR or robots?</b.

Levy-Tzedek's laboratory is also researching rehabilitation under virtual reality conditions. "We investigated people's preferences in cognitive training. Today, it's done using software that shows you, for example, a series of different colored shapes, and you have to click when you see a certain sequence. So, we divided the subjects into three groups. One executed the task like that, the second received instruction from a social robot - including feedback, conversation, a meditation break, and empathic comments like, 'Yes, that exercise was tough'. The third group was given a VR simulation in which they flew in a plane, went diving, and rode a motorcycle, in a game where they had to collect the shapes.

"In previous experiments we conducted comparing virtual reality with robots, the robots usually won, but we suspected that might have happened because virtual reality didn’t get to show its true capabilities. We wanted to let each tool demonstrate maximum capabilities, then compare. And in fact, in this experiment we received opposite results: 66% preferred virtual reality, even if there were also complaints, such as about nausea, or the helmet being uncomfortable. We think that, in the future, it may be possible to combine elements, sometimes robots, other times virtual reality, to keep things interesting."

Today, Levy-Tzedek is designing a robot for patients who have suffered from diseases other than stroke. "Our lab team comes from different fields - clinical, engineering, psychology and cognition. Our partners come from education and philosophy. I also believe in cooperative planning. When we design a robot, we include the patients and their families from the start. We ask clinicians what they observe in their clinics, what's important, and through this ongoing conversation with all stakeholders, we build very positive interactions."

Published by Globes, Israel business news - en.globes.co.il - on March 13, 2022.

© Copyright of Globes Publisher Itonut (1983) Ltd., 2022.

Prof. Shelly Levy-Tzedek  credit: Dani Machlis - Ben Gurion University of the Negev
Prof. Shelly Levy-Tzedek credit: Dani Machlis - Ben Gurion University of the Negev
Twitter Facebook Linkedin RSS Newsletters גלובס Israel Business Conference 2018