Using 3D load planning in logistics helps maximize available space, reduce operational costs, and improve operational efficiencies. Traditional load planning methods cost more because of poor weight distribution and increased fuel costs; cargo is not arranged efficiently. While there have been attempts by researchers to improve heuristics, meta-heuristics, and even Artificial Intelligence (AI) to solve 3D Bin Packing Problems (3D-BPP), the developed models do not address model adaptability, real-world, or real-world cargo constraints, such as the cargo being fragile. Additionally, there may be a priority for a specific pack. This study's goal is to develop a model of dynamic optimization for the 3D load planning process in real time by integrating reinforcement learning with a heuristic. The model is able adapt to changing constraints regarding weight, and is able to continually improve the placement of the cargo with feedback to the load planning system. The model should be able improve the heuristics, AI, or sequential models being used in the realm of weight distribution, more so than GPU and parallel processing, which should improve the quality of the system. The model used in the test case demonstrated a space utilization of 87% and a weight distribution in 96% of the test cases. The model should also improve the time to retrieve a priority case by 20% (this may be as a result of poor weight distribution). The model also demonstrated and improved adaptive ability. The model also improved the ability to retrieve priority cases (this was a result of improved weight distribution). The model also demonstrated and improved adaptive ability. The findings illustrate the efficacy of the fusion of reinforcement learning with heuristic approach decision-making, creating a synergy between heuristic efficacy and artificial intelligence flexibility. Future work should aim for enhancements of scalability beyond the IoT, real-time tracking, and hybrid optimised AI-heuristic approaches for improved real-time, dynamic decision-making within dynamic logistics environments.
Jemakmun et al. (Fri,) studied this question.