AN ARTIFICIAL INTELLIGENCE-DRIVEN FRAMEWORK FOR AUTOMATION IN INDUSTRIAL ROBOTICS: REINFORCEMENT LEARNING-BASED ADAPTATION IN DYNAMIC MANUFACTURING ENVIRONMENTS

Abstract

Industrial robots often struggle to sustain throughput, quality, and operational flexibility during disturbances because reinforcement learning (RL) adaptation is implemented without adequate sensing, data pipelines, governance controls, and workforce readiness across enterprise automation systems. This study tested an AI-driven enterprise framework linking organizational and technical enablement factors to RL-based adaptation effectiveness and, in turn, to automation performance outcomes. Using a quantitative, cross-sectional, case-based design, a 5-point Likert survey was administered across cloud and enterprise manufacturing cases; after screening, N = 210 valid responses were retained (96.2% completeness; item-level missingness < 3.0%) from engineering, production, maintenance, quality, and operations roles. Independent variables were real-time sensing and integration (X1), data quality and accessibility (X2), digital twin or simulation support (X3), governance and safety readiness (X4), and human-robot collaboration readiness (X5); RL-based adaptation effectiveness (M1) acted as the mediator and automation performance (Y) as the outcome. Analyses included descriptive statistics, reliability testing, Pearson correlations, and multiple regression with multicollinearity diagnostics. Mean scores indicated generally favorable conditions (X1 M = 3.92, SD = 0.63; M1 M = 3.77, SD = 0.64; Y M = 3.85, SD = 0.60) and internal consistency (Cronbach’s α = 0.81 to 0.89). M1 correlated with enablement factors (r = 0.39 to 0.56, all p < .001) and Y correlated with M1 (r = 0.62, p < .001). The regression model predicting Y was significant (R² = 0.54; Adjusted R² = 0.52; F(6,203) = 39.80, p < .001) with collinearity within limits (VIF 1.28 to 2.11). M1 was the strongest predictor (β = 0.41, p < .001), followed by X2 (β = 0.19, p = .002), X4 (β = 0.15, p = .007), X1 (β = 0.12, p = .039), and X5 (β = 0.10, p = .046), while X3 was not significant (p = .133). Implications are that enterprises should prioritize data quality, safety governance, and human readiness; digital twins add value when integrated with data and control processes.