Integrating Reinforcement Learning and AI Agents for Adaptive Robotic Interaction and Assistance in Dementia Care

TL;DR

This work tackles data scarcity in socially assistive robotics for dementia care by introducing an open-source simulator that jointly models PLWD states and robot caregiver behavior using probabilistic dynamics and LLM-based simulations. The main approach combines a Markovian PLWD state model with PASS-guided assistance and a reinforcement learning framework for an RL-based Robot Caregiver, augmented by GPT-4o-driven perception and action execution. Key contributions include (i) a probabilistic PLWD model capturing Forgetfulness, Confusion, Anger, and Disengagement, (ii) an RL policy-learning pipeline with a defined state/action space and a reward structure emphasizing minimal effective assistance, and (iii) end-to-end LLM-enabled perception and execution modules enabling nuanced, state-aware caregiver interactions within ADLs. The study demonstrates that RL-guided, state-aware caregiving can be more effective than random strategies in simulation, underscoring the potential for AI-driven, adaptive dementia care while highlighting the need for real-world validation and ethical, domain-expert evaluation.

Abstract

This study explores a novel approach to advancing dementia care by integrating socially assistive robotics, reinforcement learning (RL), large language models (LLMs), and clinical domain expertise within a simulated environment. This integration addresses the critical challenge of limited experimental data in socially assistive robotics for dementia care, providing a dynamic simulation environment that realistically models interactions between persons living with dementia (PLWDs) and robotic caregivers. The proposed framework introduces a probabilistic model to represent the cognitive and emotional states of PLWDs, combined with an LLM-based behavior simulation to emulate their responses. We further develop and train an adaptive RL system enabling humanoid robots, such as Pepper, to deliver context-aware and personalized interactions and assistance based on PLWDs' cognitive and emotional states. The framework also generalizes to computer-based agents, highlighting its versatility. Results demonstrate that the RL system, enhanced by LLMs, effectively interprets and responds to the complex needs of PLWDs, providing tailored caregiving strategies. This research contributes to human-computer and human-robot interaction by offering a customizable AI-driven caregiving platform, advancing understanding of dementia-related challenges, and fostering collaborative innovation in assistive technologies. The proposed approach has the potential to enhance the independence and quality of life for PLWDs while alleviating caregiver burden, underscoring the transformative role of interaction-focused AI systems in dementia care.

Figures (12)

• Figure 1: Illustration of the system architecture for our two AI agents, PLWD and Robot Caregiver. The agent PLWD is built with two modules. The agent Robot Caregiver consists of three modules for perception, decision-making, and action execution, correspondingly.
• Figure 2: The flowchart illustrates the dynamic interaction between the two AI agents, a person living with dementia (PLWD) and the robot caregiver, during a daily activity scenario. The pink and orange blocks represent the simulation of PLWD and the robot caregiver, correspondingly.
• Figure 3: A high-level illustration of the state transition model for agent PLWD. Action $a_i$ could be any of the four actions, $a_0$-$a_3$.
• Figure 4: A graph illustration of how the dynamics of PLWD's status Forgetfulness is simulated. The same mechanics applied to the other three statuses, Confusion, Anger, and Disengagement.
• Figure 5: The prompt used in our PLWD behavior simulation model to generate PLWD's realistic, detailed behavior.