Comparative Study of Contact and Non-Contact Sensing Architectures for Stewart Platform Stabilization with Adaptive RL Control

Abstract

The stabilization of unstable non-linear systems, specifically the Ball and Plate System, is a benchmark problem in control theory. This research evaluates the efficacy of three distinct sensing architectures applied to a Stewart Platform: (1) Resistive Touch Screens (Contact-based), (2) Computer Vision (External Optical), and (3) a proposed high-frequency Infrared (IR) Phototransistor Array (Internal Optical). Experimental analysis reveals that while contact-based methods are cost-effective, they introduce significant mechanical damping (friction) that masks true system dynamics. Conversely, external vision systems eliminate friction but incur computational latency (>30ms). This paper demonstrates that the proposed IR Phototransistor Array offers a superior trade-off, achieving sub-millisecond response times with zero mechanical impedance. Furthermore, we propose a transition from classical PID control to Deep Reinforcement Learning (DRL) to autonomously compensate for environmental disturbances without manual tuning.