Risk-Averse Biased Human Policies in Assistive Multi-Armed Bandit Settings

TL;DR

This paper introduces an algorithm for assistive multi-armed bandit scenarios that accounts for human risk-averse biases, enabling robots to make more rational decisions and improve team utility.

Contribution

It presents a novel algorithm that models human risk aversion within multi-armed bandit frameworks, enhancing human-robot collaboration.

Findings

Algorithm effectively handles arbitrary reward functions.

Robots can mitigate human biases to improve decision-making.

Potential for broader application in assistive systems.

Abstract

Assistive multi-armed bandit problems can be used to model team situations between a human and an autonomous system like a domestic service robot. To account for human biases such as the risk-aversion described in the Cumulative Prospect Theory, the setting is expanded to using observable rewards. When robots leverage knowledge about the risk-averse human model they eliminate the bias and make more rational choices. We present an algorithm that increases the utility value of such human-robot teams. A brief evaluation indicates that arbitrary reward functions can be handled.