AI-Driven Simulation for Optimizing Warehouse Logistics

Use Case:

Evaluate and optimize inbound/outbound flow handling and stock management in a real-world distribution center; identify performance bottlenecks and derive decision-making insights from simulation scenarios.

Outcome:

• Implementation of a flexible sandbox for logistics performance analysis;
• Integrated AI engine with Surrogate, Explainable AI, and Active Learning modules;
• Prioritized simulations based on relevance, saving time and computational cost;
• Deployment-ready solution for integration.

Ecosystem Support:

• Supported under the StairwAI project;
• Mentorship provided by experts from FundingBox and the University of Bologna;
• Technical guidance and AI infrastructure provided by assigned mentors.

AI Relevance:

• Showcases how AI + simulation improves complex decision-making in industrial environments;
• Demonstrates value of hybrid AI architectures (XAI + AL) for model interpretability and resource efficiency;
• Offers a replicable, scalable approach for digital transformation in logistics and operations.

Summary:

STAM S.R.L., an Italian engineering and technology company, developed AITrawel, an AI-enhanced simulation solution to improve warehouse logistics and decision-making. Originally focused on transport optimization, the project shifted its attention to modeling and analyzing the performance of a real-world autonomous system operating in the logistics center of a major global sports retailer. AITrawel uses a simulation-based “sandbox” supported by AI modules that help analyze system behavior under varying configurations. The AI integration allows for: i) mapping initial conditions to simulation outcomes, ii) prioritizing critical simulation runs, iii) generating explainable, actionable insights.

The AI engine includes three core components: a surrogate module linking simulation inputs/outputs, an eXplainable AI module generating human-interpretable insights, an active Learning module that optimizes resource use by prioritizing valuable simulation scenarios. The system is coded in Python with Cython packaging and demonstrates clear scalability. Results from the pilot will be integrated into STAM’s workflow, allowing broader deployment across logistics, supply chain, and manufacturing use cases.