EXPLORE PUBLICATIONS BY COUNTRIES

Artificial Intelligence (AI), Machine Learning (ML), LSTM, Q-Learning, Warehouse Automation, Predictive Inventory Management, Reinforcement Learning, Trustworthy AI.

We introduce a reliable hybrid AI system for cognitive warehousing that combines predictive stockkeeping unit (SKU)-level inventory management with intelligent physical automation in a closed-loop configuration. We combine an LSTM model for SKU-level forecasting with Q-learning for routing Automated Guided Vehicles (AGVs) optimally and test the system on simulation based on a synthetic data set of 422 heterogenous SKUs with seasonality, promotions, and mixed lead times. LSTM forecaster controls an inventory optimization engine (dynamic slotting, safety stock, and reorder points), whose recommendations are executed by a warehouse execution system which navigates AGVs with learned navigation policies. The proposed framework outperforms baseline statistical forecasts and non-learning path planners in terms of reducing order-fulfillment time and picker travel distance, improving picking accuracy, decreasing stockouts and holding cost, and showing quantifiable operational improvement. In line with U.S. NSTC guidelines for human-focused and moral AI, the architecture puts highest emphasis on (i) interpretability (transparency of KPIs and auditable decisioning), (ii) resilience (stress testing on demand variation and congestion), and (iii) human control (policy safeguarding and operator regulation). We introduce simulation-only evaluation constraints and traceout paths for pilot rollout, e.g., extensions to Deep Q-Networks/Proximal Policy Optimization and computer vision-based quality inspection. The findings suggest that closed-loop coupling of predictive analytics and learned control can provide reliable, scalable gains in warehouse productivity and inventory health.