An Adaptive Behaviour-Based Strategy for SARs interacting with Older Adults with MCI during a Serious Game Scenario

TL;DR

This paper presents an adaptive reinforcement learning-based strategy for Socially Assistive Robots to maintain engagement of older adults with MCI during serious game-based cognitive training, addressing monotony and dropout issues.

Contribution

It introduces a novel reinforcement learning approach for adaptive robot behavior to enhance engagement in cognitive training for older adults with MCI.

Findings

Reinforcement learning algorithm converged successfully.

Adaptive strategy improved engagement levels.

Simulation results validated the approach's effectiveness.

Abstract

The monotonous nature of repetitive cognitive training may cause losing interest in it and dropping out by older adults. This study introduces an adaptive technique that enables a Socially Assistive Robot (SAR) to select the most appropriate actions to maintain the engagement level of older adults while they play the serious game in cognitive training. The goal is to develop an adaptation strategy for changing the robot's behaviour that uses reinforcement learning to encourage the user to remain engaged. A reinforcement learning algorithm was implemented to determine the most effective adaptation strategy for the robot's actions, encompassing verbal and nonverbal interactions. The simulation results demonstrate that the learning algorithm achieved convergence and offers promising evidence to validate the strategy's effectiveness.