Bias and Performance Disparities in Reinforcement Learning for Human-Robot Interaction

Zoe Evans; Matteo Leonetti; Martim Brandão

Bias and Performance Disparities in Reinforcement Learning for Human-Robot Interaction

Zoe Evans^*, Matteo Leonetti, Martim Brandão

^*Corresponding author for this work

Informatics

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

91 Downloads (Pure)

Abstract

Bias has been shown to be a pervasive problem in machine learning, with severe and unanticipated consequences, for example in the form of algorithm performance disparities across social groups. In this paper, we investigate and characterise how similar issues may arise in Reinforcement Learning (RL) for Human-Robot Interaction (HRI), with the intent of averting the same ramifications. Using an assistive robotics simulation as a case study, we show that RL for HRI can perform differently across models with different waist circumferences. We show this behaviour can arise due to representation bias - unbalanced exposure during training - but also due to inherent task properties that may make assistance difficult depending on physical characteristics. The findings underscore the need to address bias in RL for HRI. We conclude with a discussion of potential practical solutions, their consequences and limitations, and avenues for future research.

Original language	English
Title of host publication	2025 ACM/IEEE International Conference on Human-Robot Interaction
Publication status	Accepted/In press - 2025

Access to Document

hri2025_assistive_gym_auditAccepted author manuscript, 556 KBLicence: CC BY

Cite this

@inbook{8611b6e3911b4bc2a05df057779c681d,

title = "Bias and Performance Disparities in Reinforcement Learning for Human-Robot Interaction",

abstract = "Bias has been shown to be a pervasive problem in machine learning, with severe and unanticipated consequences, for example in the form of algorithm performance disparities across social groups. In this paper, we investigate and characterise how similar issues may arise in Reinforcement Learning (RL) for Human-Robot Interaction (HRI), with the intent of averting the same ramifications. Using an assistive robotics simulation as a case study, we show that RL for HRI can perform differently across models with different waist circumferences. We show this behaviour can arise due to representation bias - unbalanced exposure during training - but also due to inherent task properties that may make assistance difficult depending on physical characteristics. The findings underscore the need to address bias in RL for HRI. We conclude with a discussion of potential practical solutions, their consequences and limitations, and avenues for future research.",

author = "Zoe Evans and Matteo Leonetti and Martim Brand{\~a}o",

year = "2025",

language = "English",

booktitle = "2025 ACM/IEEE International Conference on Human-Robot Interaction",

}

TY - CHAP

T1 - Bias and Performance Disparities in Reinforcement Learning for Human-Robot Interaction

AU - Evans, Zoe

AU - Leonetti, Matteo

AU - Brandão, Martim

PY - 2025

Y1 - 2025

N2 - Bias has been shown to be a pervasive problem in machine learning, with severe and unanticipated consequences, for example in the form of algorithm performance disparities across social groups. In this paper, we investigate and characterise how similar issues may arise in Reinforcement Learning (RL) for Human-Robot Interaction (HRI), with the intent of averting the same ramifications. Using an assistive robotics simulation as a case study, we show that RL for HRI can perform differently across models with different waist circumferences. We show this behaviour can arise due to representation bias - unbalanced exposure during training - but also due to inherent task properties that may make assistance difficult depending on physical characteristics. The findings underscore the need to address bias in RL for HRI. We conclude with a discussion of potential practical solutions, their consequences and limitations, and avenues for future research.

AB - Bias has been shown to be a pervasive problem in machine learning, with severe and unanticipated consequences, for example in the form of algorithm performance disparities across social groups. In this paper, we investigate and characterise how similar issues may arise in Reinforcement Learning (RL) for Human-Robot Interaction (HRI), with the intent of averting the same ramifications. Using an assistive robotics simulation as a case study, we show that RL for HRI can perform differently across models with different waist circumferences. We show this behaviour can arise due to representation bias - unbalanced exposure during training - but also due to inherent task properties that may make assistance difficult depending on physical characteristics. The findings underscore the need to address bias in RL for HRI. We conclude with a discussion of potential practical solutions, their consequences and limitations, and avenues for future research.

M3 - Conference paper

BT - 2025 ACM/IEEE International Conference on Human-Robot Interaction

ER -

Bias and Performance Disparities in Reinforcement Learning for Human-Robot Interaction

Abstract

Access to Document

Fingerprint

Cite this