The Bear and the Nurse: A Healthcare Robot Story That Started With the Right Question

The Bear and the Nurse: A Healthcare Robot Story That Started With the Right Question

When engineers design a hospital robot, they start with what the robot can do. When nurses design one, they start with what is driving them crazy.

That sounds obvious. It is not what the industry has been doing.

A new study from Cornell Tech, presented at the ACM CHI 2026 conference on human-computer interaction, tried a different approach. Over 14 weeks, 22 people — nurses, doctors, long-term care residents, artists, engineers, computer scientists, and craftsmen — met weekly in Cornell's MakerLAB to imagine and physically build robots for three healthcare settings: an emergency department, a sleep disorder clinic, and a long-term rehabilitation facility. They started by asking what frustrated healthcare workers and stressed patients. They built from there.

The result was not a better demo. It was a different set of outputs.

In the emergency department, the team led by nurses designed a bear-shaped robot to deliver medical kits to patient rooms before a doctor arrives. An engineer would not have pitched that. Neither would a patient advocate, a hospital administrator, or a focus group. A nurse who has watched the supply chain break down in a high-pressure resuscitation scenario pitched it. The form followed the function that nobody had articulated until someone asked.

This is the finding that matters most from the Cornell study, and it is almost entirely missing from how the healthcare robot industry talks about its own work.

The field has spent years showing what robots can do in hospitals — delivery, disinfection, rehabilitation assistance, emotional support. The demos are real. The adoption is slow. Robots get purchased, positioned in hallways, and quietly underused. Clinicians work around them. The most common explanation is that nurses and doctors need more training.

The Cornell study suggests a different diagnosis: the training gap is real, but it follows a design gap. The robots were not built around the work. The work was supposed to adapt to the robots.

"Many healthcare facilities experience challenges managing and caring for patients, yet limited research explores the common issues faced by healthcare workers and patients, and how robot design could help," said Angelique Taylor, the Andrew H. and Ann R. Tisch Assistant Professor at Cornell Tech and one of the paper's lead authors.

The Cornell team calls its output considerate embodied AI — robots attentive to social norms, spatial constraints, and human needs, not just efficiency metrics. The paper proposes eight guidelines for replicating the approach. They range from grounding designs in actual workflow constraints to building what the researchers call embodied literacy among non-technical participants, helping nurses and patients understand enough about how robots move and sense to have genuine opinions about what the machines should do.

One of the more striking findings was that physical prototyping revealed problems that interviews and sketches missed entirely. When teams moved from storyboards to full-scale cardboard mockups to working robot prototypes, issues surfaced: hallway width, patient comfort around a machine's noise profile, hygiene requirements in an emergency setting. These constraints only become visible when someone has to actually fit the robot into the room.

"We found that robots as embodied AI systems must be attuned to environmental context," Taylor said. "That means understanding the physical constraints of the built environment and patient conditions."

The MakerLAB environment played an underappreciated role. Co-author Niti Parikh, who runs the MakerLAB and founded a research collective called CRAFT@Large, described digital fabrication not as a manufacturing step but as an instrument of thought — a way for non-technical participants to move from passive observers to active shapers of what the technology would actually do.

Artists shaped how the robots looked and felt. Long-term care residents flagged moments where a machine might feel intrusive rather than helpful. Healthcare workers identified workflow frictions that technologists routinely overlook. The hierarchy flattened in the MakerLAB, which Parikh described as a neutral third place separate from clinical hierarchies and academic pressure, and the designs improved as a result.

The three settings produced meaningfully different robots, which is itself a finding. The sleep clinic team built a gentle, concierge-style robot with calming lights to guide patients through unfamiliar nighttime procedures. The long-term rehabilitation team focused on social connection — entertainment, scheduling, something to reduce the isolation that defines too many institutional care settings. The robots were not one-size-fits-all solutions because the co-design process did not treat them as one.

Across all three settings, the robots worked best when handling repetitive, non-clinical tasks — the kind of work that consumes time and attention without requiring judgment, empathy, or human connection. The implication is not that robots will replace what nurses do. It is that robots might finally stop getting in the way of it.

The study has limits. This was one 14-week workshop at Cornell Tech. The eight guidelines have not been validated in actual hospital deployments. A co-designed prototype is not the same as a deployed system, and the MakerLAB's hierarchy-free atmosphere may not map cleanly onto the real pressures of a functioning clinical environment. The robots that emerged from this process are proposals, not products.

But the methodology is portable, and the central lesson is not new. It is old enough that it should have already been the standard. Ask the people who do the work what the work needs. Build from there. The bear-shaped robot did not come from a technology breakthrough. It came from a nurse being in the room when the question was being asked.

The healthcare robot industry has no shortage of impressive demos. What it has is a shortage of robots that nurses actually want to work next to. The Cornell study suggests the gap is not hard to fix. It just requires asking the right question first.