Active Health Data Collection Using Social Robot

This paper aims to modernize the manner in which patient reported outcome measures (PROM) questionnaires are conducted using simplistic reinforcement learning (RL) techniques. The solution proposed in this paper consists in using an autonomous robot to perform the PROM questionnaires. In this paper the viability of two RL algorithms will be studied regarding the ability to learn and exploit the patients' daily routine in order to determine the best possible moment to interact with the user and perform the PROM to obtain a completely answered questionnaire. The main problem behind PROMs is the lack of completion of these questionnaires, using an autonomous agent to conduct the PROMs we aim to increase the user's engagement throughout the treatment and increase the number of completed questionnaires. In order to study the performance of the proposed RL algorithms a simulation environment was created to simulate the house environment of the user and the user's availability to answer a questionnaire. Throughout this study several approaches were considered regarding the state definition, reward definition and working horizon. Through the permutation of these different approaches the best configuration for each of the two proposed RL algorithms was determined to solve to the problem at hand. This work showed that using RL to solve this problem was a viable solution when it came to determine the best moment for interacting with the user for a specific day.