Automated pallet loading of irregularly shaped objects: A deep reinforcement learning and multi-criteria optimization method

Palletizing in manufacturing presents a formidable challenge, particularly when dealing with irregularly shaped objects. This paper introduces a novel approach to optimizing pallet loading scenarios integrating Deep Reinforcement Learning (DRL) and heuristic methods with stability assessment and constraint satisfaction within an automated palletization pipeline. The proposed solution consists of four key components. First, each object undergoes preprocessing, involving shape extraction from data files and the generation of metrics to evaluate stability and palletization suitability. Subsequently, object selection is performed using either a DRL agent—trained to predict optimal loading sequences—or a rule-based prioritization strategy, enabling a comparative analysis of selection methods. Constraint satisfaction techniques are then applied to narrow down the search space for candidate placement positions. Finally, optimal object placement is determined using a Multi-Criteria Decision-Making (MCDM) approach that evaluates candidate positions and orientations based on weighted performance criteria. The proposed framework is validated through a case study in the architectural aluminum industry, demonstrating its pivotal role in automating a production line responsible for sorting and packaging customer orders. • Complementary integration of DRL and heuristic optimization for pallet loading of irregularly shaped objects. • Physics-informed optimization incorporating stability assessment for secure stacking. • Comparative evaluation of DRL-based versus rule-based object selection methods. • Industrial validation addressing manual and error-prone palletizing processes. • Modular, scalable system with automated data extraction, constraint satisfaction, and multi-criteria placement.