BoxStacker: Deep Reinforcement Learning for 3D Bin Packing Problem in Virtual Environment of Logistics Systems

Area of Science:

• Robotics and Automation
• Artificial Intelligence
• Operations Research

Background:

• Manufacturing systems require enhanced resilience and self-organization to manage supply chain disruptions.
• Increasing automation in logistics necessitates advanced solutions due to a shortage of human experts.
• Deep Reinforcement Learning (DRL) offers a powerful approach for complex problem-solving by integrating artificial neural networks.

Purpose of the Study:

• To investigate the application of Deep Reinforcement Learning for real-time sequential 3D bin packing.
• To leverage a game engine for intuitive visualization and realistic training of DRL models in logistics.
• To demonstrate the efficacy of DRL in addressing dynamic logistical challenges.

Main Methods:

• Utilized a game engine integrated with a physical engine for a realistic simulation environment.
• Trained a Deep Reinforcement Learning model to solve the sequential 3D bin packing problem.
• Employed visualization tools within the game engine for intuitive observation of the learning and results.

Main Results:

• The Deep Reinforcement Learning model successfully addressed the real-time sequential 3D bin packing problem.
• The game engine provided an intuitive and realistic environment for DRL training and validation.
• The proposed approach demonstrated effectiveness in dynamic logistical settings.

Conclusions:

• Deep Reinforcement Learning, combined with game engine visualization, is a promising solution for complex, real-time logistical problems.
• This methodology enhances the adaptability and self-organization of manufacturing systems.
• The approach holds significant potential for optimizing automated logistics in dynamic environments.