The Influence of Feedback Type in Robot-Assisted Training

Neziha Akalin, Annica Kristoffersson, Amy LoutfiMultimodal Technol. Interact 2019 · DOI

TL;DR

This study investigates how different types of feedback (flattering, positive, and negative) from a social robot affect older adults' experience during physical and cognitive training. The findings suggest that flattering and positive feedback lead to higher acceptance, better attitudes towards technology, and increased intention to use the robot, while negative feedback surprisingly made the robot seem easier to use.

What Problem Does This Paper Address?

As our global population ages, there's a growing need for effective and scalable interventions to maintain the physical and cognitive health of older adults. Traditional healthcare approaches are resource-intensive, making socially assistive robots (SARs) a promising avenue for providing assistance, particularly in training and rehabilitation. However, the effectiveness of SARs hinges on how users perceive and interact with them. This paper specifically addresses the crucial role of feedback in human-robot interaction (HRI) within the context of robot-assisted training for older adults. While previous research has explored feedback, this study aims to isolate and examine the impact of different feedback types on various aspects of the user experience, including acceptance, safety, attitude, and performance, in a specific training scenario for this demographic.

Methods

The researchers conducted a user study with 23 older adults (aged 66-88) who performed a physical and cognitive training task involving remembering and repeating arm motions guided by a Pepper robot. Participants were randomly assigned to one of three feedback conditions: flattering, positive, or negative. The training involved three difficulty levels, and the robot provided feedback after each exercise.

To measure the impact of feedback, participants completed several questionnaires:

  • Pre-experiment: Demographics, SF-12 Health Survey, and an Attitude Towards Robots Questionnaire.
  • Post-experiment: A modified Almere Robot Acceptance Questionnaire, a Sense of Safety and Security Questionnaire, a Task Evaluation Questionnaire, and the Attitude Towards Robots Questionnaire again.

The study also tracked participants' performance by recording and analyzing their arm motions. Statistical analyses, including one-way ANOVAs and Tukey HSD post-hoc tests, were used to compare the groups.

Datasets

The study did not introduce new datasets for training or evaluation. Instead, it utilized participant performance data collected during the experiment for analysis. The arm exercise recognition model was trained on a self-built dataset of RGB videos of participants performing six different arm motions.

Key Insights and Ideas

  • Flattery Works (for Older Adults): Contrary to what one might expect, flattering feedback was not only well-received but also led to higher ratings in key areas like attitude towards technology, intention to use, and perceived usefulness. This suggests that for this demographic, positive reinforcement, even if exaggerated, can significantly boost engagement and acceptance.
  • Negative Feedback's Double-Edged Sword: While negative feedback generally led to lower scores in acceptance and attitude, it surprisingly resulted in participants perceiving the robot as easier to use. This could be due to a tendency to anthropomorphize the robot or a perceived increase in its "agency" when it provides critical feedback.
  • Safety and Security are Crucial for Acceptance: The study found a moderate positive correlation between robot acceptance and the sense of safety and security. This highlights that feeling safe and secure around a robot is a fundamental requirement for its successful adoption, especially for older adults.
  • Performance Doesn't Always Align with Feedback: Interestingly, there was no statistically significant difference in actual task performance across the feedback groups. However, the flattering feedback group showed the highest average performance, suggesting a potential indirect effect of positive feedback on motivation or effort, even if not on objective accuracy.
  • Social Actor Perception: Participants in the flattering and positive feedback groups tended to treat the robot as a "social actor," displaying politeness and even attempting to imitate the robot's movements. This aligns with the Computer as Social Actor (CASA) paradigm.

Key Results

  • Robot Acceptance: Participants in the flattering and positive feedback groups rated constructs like Attitude Towards Technology, Intention To Use, Perceived Usefulness, and Social Influence significantly higher than those in the negative feedback group.
  • Task Evaluation: Participants in the flattering feedback group liked the training more and perceived the robot as a more appropriate training coach compared to the negative feedback group.
  • Sense of Safety and Security: Flattering feedback positively influenced how safe and calming participants perceived the robot to be, with significant differences compared to the negative feedback group.
  • Performance: While no statistically significant difference in performance was found, the flattering feedback group had the highest average performance, followed by the positive feedback group.

Limitations and Future Work

The authors acknowledge several limitations:

  • Novelty Effect: Most participants interacted with the Pepper robot for the first time, meaning the study was subject to a novelty effect, which could influence initial perceptions.
  • Short-Term Study: The experiment was a short-term interaction. Future work should involve long-term trials to observe how acceptance and perceptions change over time.
  • Participant Demographics: The study participants were all Swedish and generally healthy and independent. Future research should include cross-cultural studies and investigate interactions with more dependent older adults.

The authors suggest future work should:

  • Investigate the relationship between interaction-related objective aspects (like facial features) and subjective evaluations.
  • Vary the level of difficulty more dynamically to better fit individual participants.
  • Avoid negative verbal feedback in future robot designs for older adults.

Summary updated

BibTeX
@article{akalin2019influence,
  author = {Neziha Akalin and Annica Kristoffersson and Amy Loutfi},
  title = {{The Influence of Feedback Type in Robot-Assisted Training}},
  journal = {Multimodal Technol. Interact.},
  year = {2019},
  doi = {10.3390/mti3040067},
}
Figure