Do robots need bodies? UMSI researchers say it depends

When people imagine a robot, they often picture a machine that moves through the world, shares physical space with humans and acts on its surroundings. 

But in human-robot interaction research, robots do not always appear in physical form. Researchers frequently study human responses to robots without using a physical robot in the room. Instead, participants may watch a video of a robot, interact with a simulated robot on a screen or encounter a robot in virtual reality. 

These methods can make studies cheaper, safer and easier to scale, but they also raise a central question for the field: Do people respond to these virtual representations the same way they would respond to a robot which is sharing physical space with them?

A new paper co-authored by University of Michigan School of Information researchers argues the field may be asking the wrong question.

Rather than debating whether virtual robots are valid stand-ins for physically embodied robots, the authors say researchers should ask when, where and for whom virtual representations are valid.

“This topic stems from a methodological tension in the field,” said Lionel P. Robert Jr., professor of information at UMSI and professor of robotics at U-M College of Engineering. “Many researchers employ virtual representations of robots, like simulations or VR, when conducting research. This is done for a number of reasons: cost, safety and sample size. However, many researchers question the validity of this approach.”

The paper, “Embodied or Virtually Represented: Navigating the Embodiment Debate in Human-Robot Interaction,” was co-authored by UMSI graduate Connor Esterwood and assistant professor at Wayne State University and Xin Ye, Ruijia Guan and Robert of UMSI. 

The debate centers on what researchers call the “embodiment hypothesis,” or the idea that a robot’s physical presence shapes how humans experience, understand and interact with it. Some studies have found that physically present robots are perceived as easier to use, more useful or more engaging than virtual alternatives. Other studies, however, have found little meaningful difference in how people accept or respond to physically embodied versus virtually represented robots.

“Many researchers assume there should be a difference, but the fact that we do not always see it suggests we do not understand why embodiment matters,” Robert said. “Taking a step back and theoretically engaging in exactly why and when embodiment matters would help us understand what is really driving the phenomenon of interest.”

That distinction matters as robots become more common in workplaces, health care, education and public life. A physical robot may be essential for studying tasks that involve handoffs, movement through space, safety or physical risk. For example, a person’s comfort around a robotic arm in an industrial setting may depend heavily on whether that arm is actually present in the room.

But for other questions, such as how people respond to a robot’s verbal cues, persuasive communication or social signals, a virtual representation may be enough.

“According to the embodiment hypothesis, physical presence matters because bodily engagement fundamentally shapes human experience, cognition, social connection and social engagement,” Robert said. “But physical presence may not matter for tasks focused on verbal cues, persuasive communication or general social perception. We really do not know yet, and we should start trying to find out.”

The authors argue that this approach could make human-robot interaction research more accessible and more rigorous. Virtual robots can allow researchers to recruit larger and more diverse participant pools, reduce costs and test ideas in safer environments before moving into real-world settings.

It could also help researchers and designers better understand where physical robots are necessary and where virtual robots, agents or simulations can achieve similar results.

“By recognizing that virtual representations are valid for certain tasks, researchers can match the tool to the task,” Robert said. “Using virtual robots allows institutions with limited funding or infrastructure to participate in research, supporting larger and more diverse study samples.”

RELATED 

Read “Embodied or Virtually Represented: Navigating the Embodiment Debate in Human-Robot Interaction,” in Science Robotics. 

Learn more about Lionel Robert’s work on human-robot interaction by visiting his UMSI faculty profile.