Affective Agent Brings Personalized Intervention Reasoning to Wearable Devices

Summary

Affective Agent is a three-layer reference architecture for deciding whether, when, and how a wearable system should intervene under uncertainty. It combines a compact language model with physiological evidence, situational context, and user history on wearable-class hardware. The system is designed to operate without a cloud connection or per-user retraining. Its perception, personalization, and reasoning layers adapt to individuals through host-managed structured memory rather than user-specific weight updates. The authors instantiate the architecture for indoor environmental quality control and evaluate it on held-out, simulator-generated longitudinal scenarios. These scenarios vary physiological state, context, signal quality, and prior intervention history. The evaluation reports that memory-based personalization and a two-pass structured reasoning process improve intervention decisions within this synthetic setting. The work presents on-device decision-making as a possible step from wearable state inference toward closed-loop personalized intervention, while the reported evidence remains limited to the simulator-based evaluation described in the paper.