1 of 10

Impacts of Robot Learning on User Attitude and Behavior

Nina Moorman, Erin Hedlund-Botti, Mariah Schrum,

Manisha Natarajan, Matthew C. Gombolay

2 of 10

In-home care robots can provide support and enable CEP members to maintain independence longer.

3 of 10

Low Involvement High Involvement

How does robot learning impact robot perceptions?

Download

Reinforcement

Learning

Interactive

Reinforcement

Learning

Learning from

Demonstration

Robot learning is useful and will be visible

Because this robot learning will likely be visible to the end-user, and can even involve the user, the way in which the robot learns may have an impact on the end-user’s perception of the robot.

Different learning methods will require the robot to interact with the environment and the end-user in different ways. For example, some learning methods will require the robot to explore the environment and fail many times before succeeding. Similarly, some learning methods will require the end-user to directly interact with the robot.

These differences between robot learning methods may alter an end-user’s perception of the robot and affect adoption. We investigate how observing robot learning impacts an end-user’s perception of a robot. The four learning conditions are as follows. In the download condition, the robot downloads knowledge from the ``cloud`` about how to do the task. In the Reinforcement Learning condition, the robot learns the task through trial and error. In the Interactive Learning condition, again the robot learns through trial and error, but this is supplemented with feedback from a human teacher labeling the robot action as good or bad. Finally, in the Learning from Demonstration condition, the human teacher physically manipulates the robot to show it how to accomplish the task.

We investigate how the method of learning – broken into conditions representing differing degrees of user involvement in the robot learning – impacts perceptions of the robot.

4 of 10

Study Design

Robot Learning

Robot Testing

Human Intervention

5 of 10

Research Questions

RQ1) How does robot learning method impact a user’s perception of the robot?

RQ2) Do robot perceptions differ between the general population and care partners?

RQ3) What role does robot embodiment play on robot perceptions?

6 of 10

How does robot learning method impact a user’s perception of the robot?

High user-involvement learning conditions corresponded to lower perceived anthropomorphism

Low Involvement High Involvement

We first investigate how robot learning method impacts a user’s perception of the robot. We compare how human-like, or anthropomorphic, the robot is perceived to be when participants observe the robot learn in our four robot learning conditions.

We find that observing the robot learning impacts perceived anthropomorphism (𝑝 = .002), and that higher user involvement in robot learning corresponds to lower anthropomorphism. This means that robots are perceived as being more human-like if their learning has low user involvement. Since prior work has shown that anthropomorphism is correlated with various other perceptions of the robot such as trust, this finding suggests that researchers should consider how anthropomorphism might play a role in their study when choosing the method of robot learning

7 of 10

Do robot perceptions differ between the general population and care partners?

Care partners reported lower perceived robot success on the medicine subtask.

General Population

Care Partners

8 of 10

What role does robot embodiment play on robot perceptions?

Remote

In Person

Physically present robots are more favorably perceived.

COVID-19 limits our in-person interactions with care members and care partners. Therefore, for this study, we choose to conduct our study with care partners, and to do so remotely. However, prior work has shown physical presence to impact user perception [48]. In order to avoid confounds of care partners being remote and the general population being in person, we evaluate the general population both remotely and in person.

We find that in-person participants report more favorably than remote participants on a variety of subjective metrics including perceived safety (𝑝 < .001), reliability (𝑝 < .001), ease of use (𝑝 = .005), attitude (𝑝 < .001), intent to use (𝑝 = .014), and trust in the robot (𝑝 < .001). This highlights the need for researchers to consider physical presence in study design.

9 of 10

Ongoing work

We are currently working with CEP members to understand

what kinds of supports is desired from a robot

how members might approach teaching robots these preferred tasks

10 of 10

Thank You!

ninamoorman@gatech.edu

The Cognitive Empowerment Program is supported by Emory’s Brain Health Center, Georgia Institute of Technology, and generous investments from Cox Enterprises, Inc. and the James M. Cox Foundation.