Prospect Theory in Physical Human-Robot Interaction: A Pilot Study of Probability Perception

TL;DR

This study investigates how people perceive and respond to probabilistic robot actions in a physically coupled task. It uses a two-action robot (assist vs disturb) with varying disturbance probabilities and records force inputs to identify behavioral strategies. Bayesian logistic regression and cumulative prospect theory models are fitted to participants’ decisions, revealing two clusters: an always-compensate group and a trade-off group with probability-weighted risk perception. The findings suggest CPT-inspired, personalized decision models can inform adaptive, safer pHRI controllers, though further work with larger samples is needed.

Abstract

Understanding how humans respond to uncertainty is critical for designing safe and effective physical human-robot interaction (pHRI), as physically working with robots introduces multiple sources of uncertainty, including trust, comfort, and perceived safety. Conventional pHRI control frameworks typically build on optimal control theory, which assumes that human actions minimize a cost function; however, human behavior under uncertainty often departs from such optimal patterns. To address this gap, additional understanding of human behavior under uncertainty is needed. This pilot study implemented a physically coupled target-reaching task in which the robot delivered assistance or disturbances with systematically varied probabilities (10\% to 90\%). Analysis of participants' force inputs and decision-making strategies revealed two distinct behavioral clusters: a "trade-off" group that modulated their physical responses according to disturbance likelihood, and an "always-compensate" group characterized by strong risk aversion irrespective of probability. These findings provide empirical evidence that human decision-making in pHRI is highly individualized and that the perception of probability can differ to its true value. Accordingly, the study highlights the need for more interpretable behavioral models, such as cumulative prospect theory (CPT), to more accurately capture these behaviors and inform the design of future adaptive robot controllers.

Figures (6)

• Figure 1: Experimental setup. (a) Based on the information displayed on the monitor, the participant physically interacts with the ArmMotus M2 robot via the handle. (b) The monitor presents the game status and target area. Information about the robot’s behavior is shown on the left information panel. The green cursor provides real-time feedback on the robot’s handle position. A “3, 2, 1, Go!” countdown signals the start of each trial.
• Figure 2: The experiment protocol. The experiment was comprised of three phases: familiarization, training and main experiment. The main experiment phase was then itself divided into two rounds each containing 5 blocks for which all trials had a fixed perturbation probability. A single block was completed only after the participant successfully executed 10 reaches.
• Figure 3: Participant compensation probabilities across different robot perturbation probabilities. For each participant, results from the first round are shown with solid lines, and results from the second round with dashed lines. It is noted that some participants exhibited exactly the same action pattern in their second rounds, causing the corresponding dashed lines to coincide.
• Figure 4: Model fitting results to map perturbation probability to compensation probability. Blue dots denote the raw data from each round, while the red and green line show the BLR and CPT fits, respectively.
• Figure 5: Distribution of the logistic regression fitting results for all participants. The red boxes highlight the two distinct clusters: A-"always-compensate" strategy; T-"trade-off" strategy.