Mechanical Arm Packing-Unpacking Control for 3D Bin Packing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing algorithms for 3D-BPP focus primarily on packing items without considering the crucial human experience of unpacking, which is essential for optimizing space utilization in logistics packing scenarios, leading to suboptimal performance.
Innovation Solution
A method and system for cooperating packing and unpacking actions using a packing-unpacking network (PUN) based on deep reinforcement learning, incorporating packing and unpacking heuristic algorithms to learn optimal strategies for mechanical arms, enhancing space utilization by removing improperly placed items and placing them back in the product bin.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional heuristic packing algorithms are used, then the packing process is simple and fast, but the space utilization rate is low because unpacking actions are not considered
Solution Approach 1:
The patent introduces unpacking actions as the inverse of packing actions. The mechanical arm can remove items from the product bin and place them in a buffer area, then later repack them. This inversion allows the system to correct suboptimal packing decisions and achieve better space utilization while maintaining packing efficiency through the learned policy.
2Reliability
If DRL-based packing algorithms are used, then the packing strategy has better robustness and generalization, but the algorithm complexity increases and computational resources are consumed
Solution Approach 1:
The patent transforms the continuous packing problem into a discrete decision-making process by defining specific actions (pack at position, unpack from position, move from buffer to position) and states (item positions, bin occupancy). This parameter discretization enables the use of DRL while controlling computational complexity through action masking and state representation optimization.
Solution Approach 2:
The patent introduces a buffer area as an intermediary between the conveyor belt and the product bin. Items can be temporarily stored in the buffer during unpacking operations, allowing the system to rearrange items for optimal packing without losing any items. This intermediary simplifies the decision space by providing a holding area that decouples immediate packing decisions from final arrangement.
3Ease of manufacture
If only packing actions are considered, then the algorithm is easier to implement, but the final space utilization is suboptimal because items cannot be rearranged
Solution Approach 1:
The patent performs preliminary unpacking actions to remove items from suboptimal positions before placing new items or rearranging existing ones. By proactively creating space through unpacking, the system can achieve better overall arrangement and space utilization while maintaining a relatively simple algorithm structure through the sequential nature of the learned policy.
Data Source
AI summary
A method for cooperating packing actions and unpacking actions of mechanical arm based on deep reinforcement learning (DRL), wherein a synergistic effect between a packing action and an unpacking action is learnt by means of a packing-unpacking network (PUN), thereby improving the packing precision and efficiency; and a packing heuristic algorithm and an unpacking heuristic algorithm are provided on the basis of human experience and are combined with the PUN, thereby further improving the packing precision and efficiency. Also provided are a system for cooperating packing actions and unpacking actions of mechanical arm based on DRL, a computer-readable storage medium, and an electronic device.