Designing social robots for cognitive impairments

01Does a social robot need to look real to work well?

Healthcare is increasingly turning to social robots to support patients, but most of that work assumes a robot has to look convincingly human, or at least convincingly alive, to be taken seriously. Abstract, minimal robots are cheaper and easier to produce than realistic ones. If a simple, abstract robot can still provide meaningful social interaction, that opens up design opportunities a realistic one can’t reach. Abstract, minimal robots are cheaper and easier to produce than realistic ones. If a simple, abstract robot can still provide meaningful social interaction, that opens up design opportunities a realistic one can’t reach. This project set out to test that assumption directly, as part of my PhD research at VU University Amsterdam.

An aging population is putting real strain on healthcare systems, and social robots are increasingly explored as an alternative form of support, especially for people with cognitive disabilities such as acquired brain injury and dementia.

That’s why this project focused specifically on people with cognitive disabilities such as acquired brain injury and dementia, designing a social robot that could support people dealing with real cognitive constraints. Constraints in realism could make the robot simpler, cheaper, and easier to produce, but would it still be accepted and trusted by the people it was designed to help? A review of low-cost social robots found that simplifying a robot’s hardware and appearance can make it more accessible to produce and deploy at scale. In 2026, more than 10 years later, this is still a relevant constraint. Source A review of low-cost social robots found that simplifying a robot’s hardware and appearance can make it more accessible to produce and deploy at scale. In 2026, more than 10 years later, this is still a relevant constraint. Source

02Polygon was designed to look like nothing in particular

To test the question cleanly, the robot itself had to avoid any obvious resemblance to a human, an animal, or an existing product; something that gave the user no preconceptions to fall back on. That became Polygon: a small, tabletop robot, about 30 by 20 by 15 centimeters, built for social interaction with people with acquired brain injury.

For this specific group, that abstraction mattered for a second reason beyond avoiding preconceptions. An unfamiliar form also reduced the risk of the robot carrying negative associations, something a design that leaned on an existing product category, medical equipment, a toy, a humanoid figure, couldn’t as easily sidestep. Patients with ABI are already navigating a loss of independence day to day, and a robot that read as clinical or infantilizing risked adding to that rather than helping with it. Robert A. Paauwe, David V. Keyson, Johan F. Hoorn, Elly A. Konijn, “Minimal Requirements of Realism in Social Robots: Designing for Patients with Acquired Brain Injury,” CHI EA ‘15, 2015. The paper frames Polygon’s abstract form as reducing both preconceptions about its function and negative associations the form itself might carry. Source Robert A. Paauwe, David V. Keyson, Johan F. Hoorn, Elly A. Konijn, “Minimal Requirements of Realism in Social Robots: Designing for Patients with Acquired Brain Injury,” CHI EA ‘15, 2015. The paper frames Polygon’s abstract form as reducing both preconceptions about its function and negative associations the form itself might carry. Source

The early design went through several rounds. Early paper prototypes tested scale and proportion. Later versions were laser-cut from MDF (medium-density fiberboard), giving Polygon a warm, approachable material quality rather than a sterile, plastic one.

The final robot could speak via remote control, blink, gaze left and right, and give a sense of breathing by retracting its legs into its body, though only blinking was switched on for this first evaluation with patients. Frank Thomas and Ollie Johnston’s “The Illusion of Life: Disney Animation” argues that a character’s motion, not its visual likeness, is what convinces an audience it’s alive. The same principle informed how Polygon moved. Source Frank Thomas and Ollie Johnston’s “The Illusion of Life: Disney Animation” argues that a character’s motion, not its visual likeness, is what convinces an audience it’s alive. The same principle informed how Polygon moved. Source

Each panel was cut flat, then dry-fit against its neighbors before anything was glued or fastened. The embodiment went through several iterations, to ensure all the electronics could fit inside, and that the robot could move its legs without getting stuck on its own body.

03Testing Polygon with people who have acquired brain injury

Three focus groups (n=34: 23 patients diagnosed with acquired brain injury, age range 25 to 67, 11 healthcare professionals) were held at two Dutch daytime activity centers to see how people responded to Polygon in a real setting. Robert A. Paauwe, David V. Keyson, Johan F. Hoorn, Elly A. Konijn, “Minimal Requirements of Realism in Social Robots: Designing for Patients with Acquired Brain Injury,” CHI EA ‘15, 2015. Source Robert A. Paauwe, David V. Keyson, Johan F. Hoorn, Elly A. Konijn, “Minimal Requirements of Realism in Social Robots: Designing for Patients with Acquired Brain Injury,” CHI EA ‘15, 2015. Source

Most participants were openly skeptical of robots going in, some calling them creepy, most saying they’d never talk to one. That shifted once Polygon was in the room: patients moved from vague, futuristic ideas to specific, grounded ones, from help finding their way outside to a reminder to get dressed correctly.

Two themes came up unprompted and repeatedly. Independence: several patients said they’d rather ask a machine for help than keep asking the people around them, describing that request as a burden on others rather than a neutral ask. Loneliness: patients who were physically but not cognitively or socially impaired, in particular, described wanting a low-stakes outlet, something to talk to without it turning into a conversation with family or care staff.

Small as it was, Polygon still drew real social and emotional reactions from patients. That result set up the next step: finding out how much further behavioral realism could be stripped away before that reaction stopped happening.

04Testing realism on its own, in experimental settings

From there, we ran two follow-up studies to find out how far realism could be pushed down. The first was a controlled experiment contrasting healthy participants with people with acquired brain injury. That study still remains unpublished, but the setup is worth sharing here, since it shows how we approached isolating the effect of realism on its own.

The second, published study took a different, more controlled route to the same question, built on a theoretical framework for how people perceive and respond to interactive, human-like characters. Robert A. Paauwe, Johan F. Hoorn, Elly A. Konijn, David V. Keyson, “Designing Robot Embodiments for Social Interaction: Affordances Topple Realism and Aesthetics,” International Journal of Social Robotics, 2015. Source Robert A. Paauwe, Johan F. Hoorn, Elly A. Konijn, David V. Keyson, “Designing Robot Embodiments for Social Interaction: Affordances Topple Realism and Aesthetics,” International Journal of Social Robotics, 2015. Source Participants (n=29, mean age 28.8, ranging from 18 to 56) each interacted with three robots built from LEGO Mindstorms, identical in every way except how realistic their form was. Each robot presented itself as a physiotherapy assistant and guided the participant through a set of exercises.

The robots’ visual fidelity was controlled by using LEGO. In this specific experiment, their representation was manipulated to be either a humanoid form, an animal, or mechanical. A structured questionnaire afterward found that how realistic a robot looked played only a modest role in how people perceived it. The stronger predictor of engagement, and whether people said they’d want to use the robot again, was its perceived affordances, what people believed the robot could do for them.

05The real impact of Polygon was the start of a social robot company

Some of this early work directly shaped the design philosophy behind Tessa, the social robot I later built at Tinybots (rebranded to Tessa Care in 2026), the company I co-founded. Tessa applies the same core idea, an intentionally simple, non-human form focused on what it can do rather than what it looks like, to supporting people with dementia.

The same research direction continued past Polygon itself. A later project I contributed to brought a social robot together with a home sensor network to support daily activities for people with dementia, extending the same minimal-realism approach into a new clinical context. Sara Casaccia et al., “Social Robot and Sensor Network in Support of Activity of Daily Living for People with Dementia,” Dementia Lab Conference, Springer, 2019. Source Sara Casaccia et al., “Social Robot and Sensor Network in Support of Activity of Daily Living for People with Dementia,” Dementia Lab Conference, Springer, 2019. Source

That finding became the starting point for Tessa, and for Tinybots. The same question Polygon raised, how much realism a robot needs before it stops feeling trustworthy, or worse, unsettling, shaped Tessa’s earliest design decisions too: material, price point, and how much personality to build in without ever attempting to look human.

By 2019, more than 500 Tessas were in active use across healthcare sites in the Netherlands, a number that kept growing after that. I started with a PhD question about realism. Tinybots became a full-time company before I could complete my dissertation. Right now, the PhD itself still waits to be finished. Tinybots became a full-time company before I could complete my dissertation. Right now, the PhD itself still waits to be finished. I ended up implementing the answer at scale, in real healthcare settings.