Quantifying human-robot interaction by developing a measure for safe collaboration

Camilleri, Antonella

Abstract

This research investigated whether human motion prediction by an assistive robot can always lead to a safe and efficient human-robot interaction. Using human motion, prediction can encourage interaction in close physical proximity and can help build trust in collaborative human-robot interactions. However, when the interaction gets more complicated due to external disturbances and the complex nature of the humans, it may lead to physical harm or interaction failures. Despite the literature raising safety issues in physically assistive robots, research regarding potential disturbances and adverse outcomes is limited. Therefore, this research considers the impact of external disturbances and their effect on the collaboration.

To address this gap, the present study investigates the consequences of external disturbances on the collaborative state of humans, particularly in the context of assistive tasks in natural living environments like care homes. A comprehensive examination of the impact of disturbances on human motion is conducted through surveys, human-robot interaction experiments, human motion recordings, and observational studies involving professionals in care homes. Two case studies are conducted to analyze different interaction scenarios and complexities, resulting in the collection of two time-series datasets capturing human movement.
The first case study focuses on human reaching movement in a shared workspace, utilizing movement primitives to predict and distinguish minor variations in the final reaching position. The second case study examines human movement during an assistive dressing task, introducing cognitive overloading and distractions to evaluate the effects of disturbances in a more complex interaction environment. Quantitative and Qualitative techniques are employed to identify differences in movement patterns during these irregularities, revealing that collaboration is hindered in the presence of disturbances.

The findings from the observations carried out in care homes contribute to further analysis of complex interaction and their requirements to provide safe physical assistance. The natural occurrence of successful, safe and efficient collaborative interactions witnessed in care homes, regardless of the vulnerability level of older adults, is examined and questioned. This leads to the belief that a measure of collaboration between humans and robots through input modalities is necessary for ensuring safety. This measure acts as an implicit constraint for the robot, particularly when there is variation between human and robot movement, especially during changes in the collaborative state of humans. It enables a more realistic evaluation of human motion prediction by directly assessing the safety of continuing collaborative movements.

The aforementioned case studies served as a foundation for further analysis of human movement as input modalities. To ensure physical safety, knowledge similar to that obtained from the second case study can be utilized as priors, represented in the form of latent spaces, to provide information about the human's collaborative state. This approach allows for a more accurate assessment of the safety of continuing collaborative movements. The core contribution of the thesis lies in leveraging the input modality of human movement as an affordance that ensures physical safety in assistive robots, incorporating knowledge about collaboration while considering the influence of environmental factors, human factors, and the human state.

Citation

Camilleri, A. Quantifying human-robot interaction by developing a measure for safe collaboration. (Thesis). University of the West of England. Retrieved from https://uwe-repository.worktribe.com/output/10883623

• Thesis Type: Thesis
• Deposit Date: Jun 26, 2023
• Publicly Available Date: Apr 12, 2024
• Public URL: https://uwe-repository.worktribe.com/output/10883623
• Award Date: Apr 12, 2024