Mixed reality-based online 3D pallet loading problem to achieve augmented intelligence in e-fulfilment processes

Abstract

Pallet loading operations support palletisation and truckload optimisation for e-fulfilment processes. Currently, the pallet loading problem is optimised offline using available cargo information, which is advantageous compared to typical freight operations but results in inefficiency when handling fragmented e-commerce orders. This research develops a mixed reality-based online pallet loading system (MROPLS) supported by deep reinforcement learning technology and online algorithms that dynamically decide cargo placements and orientations without prior information for pallet loading operations. The MROPLS proposes a 3-dimensional maximal-rectangle non-guillotine cutting strategy combined with a deep Q-network to increase space utilisation effectively. This approach is achieved using the lookahead algorithm, which predicts upcoming packages in the online pallet loading process and optimises package spatial location and orientation decision-making. We conduct simulation experiments to verify the system’s feasibility and performance by considering SF Express, DHL and Royal Mail package and ISO pallet sizes. The interaction effects between package types, pallet sizes and lookahead values were found and summarised to determine optimal system settings. With the aid of MROPLS, human intelligence in the online pallet loading process can be augmented, resulting in optimal palletisation in warehouse automation.