A Closed-Loop System Architecture for Neuro-Responsive Kinematic Façades: Non-Contact Sensing, Edge AI Fusion, and Reinforcement Learning Control

This paper presents the system architecture, signal-processing methodology, and control formulation of a neuro-responsive kinematic façade that closes a real-time feedback loop around a building occupant's own non-contact biometric signal. The system integrates a millimeter-wave (mmWave) radar sensing layer, an edge-computing inference layer, a reward-driven reinforcement-learning (RL) controller, and a parametric kinematic actuation layer implemented in Grasshopper for Rhinoceros. We detail the full computational pipeline: (i) non-contact acquisition of heart-rate variability (HRV) and electrodermal activity (EDA) proxies; (ii) a moving-average denoising and clamping stage; (iii) a weighted sensor-fusion formula that produces a single composite physiological stress index S(t); (iv) a monotonic response-mapping function that converts S(t) into kinematic panel rotation, correlated color temperature, and fresh-air flow targets; and (v) a Markov Decision Process formulation with a multi-term reward function for online policy learning. Component-level unit tests across four verified input scenarios confirm exact agreement between the analytically derived fusion output and the live Grasshopper/Python implementation. This paper reports a systems and methods contribution, together with in-silico pipeline verification; controlled human-subject validation is specified as a fully designed but not-yet-executed protocol, explicitly outside the present scope.