Encouraging User Autonomy through Robot-Mediated Intervention

Abstract

MOTIVATIONIn this paper, we focus on the question of promoting user autonomy at a healthcare task.During a robot-mediated intervention, socially assistive robot should seek to encourage users to learn skills and behaviors that will generalize and persist beyond the duration of the intervention. Treating a care-receiver as an apprentice rather than a dependent results in greater proficiency at self-management [2]. This philosophy must be incorporated into the design and implementation of robotmediated healthcare interventions in order for them to be accepted by real-world users.Our approach toward encouraging user autonomy and promoting generalized skill learning was to model the occupational therapy technique of graded cueing [1]. Graded cueing involves giving a patient the minimum required feedback while guiding them through a task. This method promotes generalized skill learning with the goal of getting the patient autonomous enough to perform the task without any help from the therapist. Our graded cueing model was previously used to improve the arm pose imitation ability of children with autism spectrum disorders by dynamically adjusting to the perceived ability level of each user [4]. APPROACHIn this study, the graded cueing model was expanded and implemented as part of a larger, multi-session nutrition education interaction between a child and a Dragonbot robot. Due to the integrative nature of the study, the model was not adaptive, but the focus was instead on the efficacy of graded cueing prompts in a more complex task.The study participants were 26 first grade children aged 5-8 years old, learning the basics of nutrition [5]. The goal of the larger study was to examine child-robot relationship-building across six interaction sessions over three weeks. The basis for interaction was the Dragonbot character teaching children about healthy food choices through a nutrition education game. Each week consisted of two sessions -the first session was the robot teaching the child a new nutrition concept, and the second session was a game played between the robot and the child to reinforce that new concept. Both sessions used cueing prompts, but the second session contained graded responses, as shown in Figure 1. During each of these 3 second-day sessions, the goal of the interaction was for the child to find all the healthy (fake, plastic) foods and place them on a plate in front of the robot. There also could be no unhealthy foods on the plate. This resulted in two sub-goals: choose all the healthy foods, and don't choose any of the unhealthy foods. These sub-goals were implemented with ranked levels of feedback for each individual food item in accordance with the graded cueing model. Feedback to remove unhealthy foods was first priority; feedback to add more healthy foods was given only after all unhealthy foods were removed from the plate.The example interaction shown in Figure 1 is from the second session of the third week, when children learned about healthy choices for sides at dinner time. The choices for ...

Enabling a Social Robot to Process Social Cues to Detect when to Help a User

“…A continuous value for need will allow the robot to choose an amount of assistance that is proportional to the amount of need (Wilson et al 2019). thus enabling the robot to provide multiple levels of assistance (Barnes and Stamper 2008;Greczek 2015;Wilson, Tickle-Degnen, and Scheutz 2016).…”

“…Helping too little can make the robot seem ineffective or unreliable, leading to a diminished trust in the robot (Langer et al 2019). Helping too much can be annoying, disrupt flow, and harm the user's autonomy (Greczek 2015; Simply waiting for when the user explicitly asks for help or makes a mistake may be insufficient and does not enable the agent to provide proactive or unsolicited assistance. The challenge lies in recognizing when the user needs and wants assistance, often through implicit cues (Görür et al 2017).…”

AppraisalCloudPCT: A Computational Model of Emotions for Socially Interactive Robots for Autistic Rehabilitation

“…Consequently, when designing computational models of emotions for socially interactive robots, especially for robots for a special group of people such as autistic children, one should take into account the social and communicative characteristics of such a group of people. Tere are four world-leading research groups with pioneering work in promoting social robots as useful tools in autism therapy, including the Kerstin Dautenhahn Group [18][19][20]), the Ayanna Howard Group [21,22], the Maja Matarić Group [23][24][25], and the Bram Vanderborght Group [26,27]. However, none of the 4 research groups have designed or applied computational models of emotions for the social robots used in their autism therapy studies.…”