OPA-Pack: Object-Property-Aware Robotic Bin Packing
Organizations: Department of Computer Science and Engineering, the Chinese University of Hong Kong · Hong Kong Centre for Logistics Robotics · School of Data Science, Lingnan University · Department of Mechanical and Automation Engineering, the Chinese University of Hong Kong
Abstract
Robotic bin packing is a fundamental manipulation problem with broad applications in e-commerce and warehouse automation. However, existing methods primarily focus on object geometry to optimize packing compactness without explicitly modeling object properties such as fragility, edibility, and chemical compatibility that humans typically consider during packing. This paper presents OPA-Pack (Object-Property-Aware Packing), a framework that incorporates object-property information into robotic bin-packing planning. The framework includes an object property recognition scheme based on retrieval-augmented generation and chain-of-thought reasoning to establish a dataset with object property annotations for 1,032 everyday objects. The framework further incorporates OPA-Net, which encodes candidate-object properties via a property embedding layer and represents the container state using fragility and avoidance heightmaps. We design a reward function and adopt a deep Q-learning scheme to train OPA-Net. Experimental results show that OPA-Pack improves avoidance accuracy from 52.3% to 95.0%, reduces load on fragile objects by 29.4%, and achieves a packing compactness of 0.413, compared with 0.420 for the baseline. In addition, we evaluate OPA-Pack on a physical platform using 70 packing sequences sampled from 100 scanned real-world objects. Results show that OPA-Pack outperforms the baseline and other packing heuristics in reducing the number of closely packed incompatible object pairs and the number of compressed fragile objects while preserving packing compactness.
Figures & tables
| Material | Density Level | Density ( ) | Fragility |
| Air | 0 | 0.0 | No |
| Bread | 1 | 0.3 | Yes |
| Biscuit | 1 | 0.4 | Yes |
| Wood | 1 | 0.4 | No |
| Cardboard | 1 | 0.6 | No |
| Wax | 2 | 0.9 | Yes |
| Soft | Sharp | Fragile | Edible | Household Chemical | Medicine | Ignition Item | Flammable Item | |
| Accuracy | 92% | 99% | 90% | 94% | 99% | 98% | 100% | 93% |
| Precision | 89% | 100% | 96% | 96% | 100% | 85% | 100% | 85% |
| Recall | 89% | 86% | 86% | 92% | 50% | 85% | 100% | 80% |
| F1-score | 89% | 92% | 90% | 94% | 67% | 85% | 100% | 83% |
| Material In | Material Out | Density Level | |
| Accuracy | 99% | 95% | 93% |
| Property Embedding | Map Encoders | Compactness | Avoid. Acc. | Load on Frag. | |||
| Average | 0.420 | 52.33% | 5.07 | ||||
| ✔ | 0.419 | 83.33% | 4.45 | ||||
| ✔ | ✔ | 0.378 | 91.50% | 4.23 | |||
| ✔ | ✔ | ✔ | 0.412 | 92.50% | 3.86 | ||
| ✔ | ✔ | ✔ | ✔ | 0.413 | 95.00 % | 3.58 | |
| Hardware & Stationery | 0.400 | 64.50% | 6.55 |
| Compactness | Avoid. Acc. | Load on Frag. | |
| With Candidate Poses | 0.470 | 91.5% | 3.79 |
| Without Candidate Poses | 0.413 | 95.0 % | 3.58 |
| Mean Accuracy | Std. Dev. | |
| Edible | 91.9% | 2.70 |
| Household Chemical | 98.8% | 0.40 |
| Medicine | 97.1% | 0.83 |
| Ignition Item | 99.8% | 0.40 |
| Flammable Item | 92.4% | 0.92 |
| Softness | 92.8% | 1.25 |
| Buffer Size | Compactness | Avoid. Acc. | Load on Frag. |
| 0.348 | 91.5% | 3.83 | |
| 0.413 | 95.0% | 3.58 | |
| 0.450 | 96.3% | 3.78 |
| Container Sizes | Compactness | Avoid. Acc. | Load on Frag. |
| 0.403 | 93.0% | 2.96 | |
| 0.413 | 95.0% | 3.58 | |
| 0.325 | 90.8% | 3.87 |
| Load on Frag. | Compactness | Avoid. Acc. | |
| = 2 | 4.38 | 0.423 | 91.00% |
| = 20 | 3.58 | 0.413 | 95.00% |
| = 200 | 3.51 | 0.344 | 92.33% |
| Avoid. Acc. | Compactness | Load on Frag. | |
| Scanned Real-World Objects | |||
| Method | Compactness | Avoid. Acc. | Load on Frag. |
| OPA-Pack | 0.621 | 93.0 % | 1.01 |
| IR-BPP (Baseline) | 0.616 | 51.0% | 1.10 |
| MINZ | 0.521 | 51.5% | 1.13 |
| HM | 0.503 | 59.5% | 1.10 |
| DBL | 0.510 | 56.0% | 1.19 |
| Accuracy | |
| Full Object Identification Module | 96.38% |
| CLIP Replaced with MobileNet-V2 | 95.28% |
| CLIP Replaced with ResNet | 92.78% |
| Without Multiple Images | 71.11% |
| Without Real Images | 55.00% |
| Method | Compactness | # Close Incomp. Pairs | # Compressed Fragile |
| OPA-Pack | 0.32 | 0.52 | 0.18 |
| IR-BPP (Baseline) | 0.32 | 0.94 | 0.68 |
| MINZ | 0.28 | 1.26 | 0.73 |
| HM | 0.29 | 1.05 | 0.88 |
| DBL | 0.28 | 1.45 | 0.74 |
Robotics
Robotics studies the design and control of machines that perceive, move, and act in physical environments.
See all 785 papers