Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Assistance-Seeking in Human-Supervised Autonomy: Role of Trust and Secondary Task Engagement (Extended Version)

Dong Hae Mangalindan, Vaibhav Srivastava

TL;DR

This study explores how robot actions, performance, and the introduction of a secondary task influence human trust and engagement and develops a human action model to define the probability of human reliance on the robot.

Abstract

Using a dual-task paradigm, we explore how robot actions, performance, and the introduction of a secondary task influence human trust and engagement. In our study, a human supervisor simultaneously engages in a target-tracking task while supervising a mobile manipulator performing an object collection task. The robot can either autonomously collect the object or ask for human assistance. The human supervisor also has the choice to rely upon or interrupt the robot. Using data from initial experiments, we model the dynamics of human trust and engagement using a linear dynamical system (LDS). Furthermore, we develop a human action model to define the probability of human reliance on the robot. Our model suggests that participants are more likely to interrupt the robot when their trust and engagement are low during high-complexity collection tasks. Using Model Predictive Control (MPC), we design an optimal assistance-seeking policy. Evaluation experiments demonstrate the superior performance of the MPC policy over the baseline policy for most participants.

Assistance-Seeking in Human-Supervised Autonomy: Role of Trust and Secondary Task Engagement (Extended Version)

TL;DR

This study explores how robot actions, performance, and the introduction of a secondary task influence human trust and engagement and develops a human action model to define the probability of human reliance on the robot.

Abstract

Using a dual-task paradigm, we explore how robot actions, performance, and the introduction of a secondary task influence human trust and engagement. In our study, a human supervisor simultaneously engages in a target-tracking task while supervising a mobile manipulator performing an object collection task. The robot can either autonomously collect the object or ask for human assistance. The human supervisor also has the choice to rely upon or interrupt the robot. Using data from initial experiments, we model the dynamics of human trust and engagement using a linear dynamical system (LDS). Furthermore, we develop a human action model to define the probability of human reliance on the robot. Our model suggests that participants are more likely to interrupt the robot when their trust and engagement are low during high-complexity collection tasks. Using Model Predictive Control (MPC), we design an optimal assistance-seeking policy. Evaluation experiments demonstrate the superior performance of the MPC policy over the baseline policy for most participants.
Paper Structure (16 sections, 2 theorems, 19 equations, 5 figures)

This paper contains 16 sections, 2 theorems, 19 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Dual Task Setup: Robot Supervision and target-tracking task
  3. Human Supervised Robotic Object Collection Task
  4. Target-tracking Task
  5. Data Collection for Modeling Human Behavior
  6. Data Summary
  7. Human Behavior Model
  8. Human Trust Dynamics
  9. Human Target-tracking Engagement Dynamics
  10. Human Action Model
  11. Design of human assistance seeking policy
  12. Expected Reward and Certainty-equivalent Dynamics
  13. MPC-based Assistance-seeking Policy
  14. Evaluation Experiment
  15. Discussion
  16. ...and 1 more sections

Key Result

Lemma 1

Given the probability of successful autonomous collection by the robot, $p^{\textup{suc}}_{C^1}$, the probabilities of human action $\mathbb{P}(a^{H}|T,G ,C^1), \ a^{H} \in \{a^{H+}, a^{H-}\}$, for $C^1 \in \{C^{\textup{1,L}}, C^{\textup{1,H}}\}$, and the probability of an object-collection trial be where is the expected reward for taking action $a^{R+}$ in $C^1$, considering possible human actio

Figures (5)

  • Figure 1: Experiment interface showing the target-tracking and supervisory task rendered on two adjacent screens.The setup uses ROS-Gazebo Gazebo and resources available in downs2022googleROSMoMa.
  • Figure 2: Human action model: Probability of reliance $\mathbb{P}(a^{H+}|T,G,C^1,a^{R+})$ conditioned on object collection task complexity $C^1$,$T$, and $F$
  • Figure 3: Human trust and target tracking engagement dynamics. Human trust and tracking task engagement modulate human action.
  • Figure 4: MPC policy: The optimal action for $C^{\textup{1,L}}$ is to collect, whereas for $C^{\textup{1,H}}$, action switches based on the trust and engagement states.
  • Figure 5: Cumulative reward (sum of collection and tracking rewards) statistics for both policies

Theorems & Definitions (4)

  • Lemma 1: Expected rewards
  • proof
  • Lemma 2: Certainty-equivalent evolution
  • proof