Multi-Agent Reinforcement Model for Emotional Regulation Coaching

Abstract

A hallmark of the autism spectrum disorder (ASD) is emotional regulation challenges, which can result in a behavioral crisis and stress for a caregiver. The research paper provides a multi-agent-based reinforcement learning (MARL) model that aims at coaching emotional control via adaptive feedback loops that depend on situational contexts. The system unites wearable physiological sensors, environmental IoT, and a cloud-based learning machine composed of inter-communicating agents that signify the child, caregiver, and system coach. The model was found to be 92 percent accurate in predicting emotional escalation, and it led to fewer false alerts by 61 percent using data from 60 children with ASD. The results indicate that the social interactivity of emotion regulation can be reflected in multi-agent architectures, offering individualized coaching in accordance with behavioral states without violating ethical AI principles.