The Sense of Agency in Assistive Robotics Using Shared Autonomy

TL;DR

This work addresses the gap that assistive robotics often optimize task performance at the expense of the user's sense of agency (SoA). It implements a shared autonomy framework with a user-controllable arbitration parameter $\alpha$, enabling real-time control over autonomous assistance during grasping tasks on a Kinova arm, and introduces an online proxy metric $\theta_d$ to monitor SoA. Results show a trade-off: higher autonomy improves trajectory quality but diminishes SoA, though near-optimal performance can be achieved with preserved SoA, and $\theta_d$ tracks SoA in real time. The findings inform design choices that balance control and efficiency in assistive devices and motivate extending the approach to populations with mobility impairments.

Abstract

Sense of agency is one factor that influences people's preferences for robot assistance and a phenomenon from cognitive science that represents the experience of control over one's environment. However, in assistive robotics literature, we often see paradigms that optimize measures like task success and cognitive load, rather than sense of agency. In fact, prior work has found that participants sometimes express a preference for paradigms, such as direct teleoperation, which do not perform well with those other metrics but give more control to the user. In this work, we focus on a subset of assistance paradigms for manipulation called shared autonomy in which the system combines control signals from the user and the automated control. We run a study to evaluate sense of agency and show that higher robot autonomy during assistance leads to improved task performance but a decreased sense of agency, indicating a potential trade-off between task performance and sense of agency. From our findings, we discuss the relation between sense of agency and optimality, and we consider a proxy metric for a component of sense of agency which might enable us to build systems that monitor and maintain sense of agency in real time.

Figures (8)

• Figure 1: A 24-participant study was run to explore questions on people's sense of agency and preferences for the amount of assistance during grasping tasks. During the study, participants controlled a 7-DOF robot arm with a joystick and set the amount of assistance they wanted with a dial. We find that higher levels of robot assistance led to decreased sense of agency among users.
• Figure 2: Many shared autonomy paradigms work by arbitrating between input commands from the user and commands from the assistive system. In this work, we enable users to set the value of the arbitration parameter $\alpha$ at any point during a task, which is inspired by the approach in jain2016approach. This framework enables us to uncover users' preferences for autonomous assistance throughout tasks and study the effect of robot automation on the user's sense of agency.
• Figure 3: Experimental tasks for assistance preference study. The hard task involves grasping a golf ball off of a golf tee, and the easy task involves grasping the side of a tall cylinder.
• Figure 4: Participants can perceive a difference in goal-directed robot automation (RQ1). The left bar at each $\Delta$ (the difference in $\alpha$ values of both trials in a round of the perception test) shows the proportion of participants who got the answer right (shown in purple). The right bar at each $\Delta$ show the proportion of participants who indicated the wrong condition (shown in orange) or incorrectly answered "equal" (shown in gray). Note that at $\Delta = 0$, the right answer is "equal."
• Figure 5: Participants' sense of agency decreases as goal-directed robot autonomy increases (RQ2). Participants' sense of agency scores at an $\alpha$ of $0$, $0.25$, $0.5$, $0.75$, and $1$ are shown. The data for "Intended" has been reverse coded. Table \ref{['soa_table']} details each sense of agency item and what plot they map to.