目录

RoboBPP: Benchmarking Robotic Online Bin Packing with Physics-based Simulation

RoboBPP is a comprehensive benchmark for Robotic Online 3D Bin Packing, designed to bridge the gap between algorithmic packing performance and physical feasibility in real robotic scenarios.
The benchmark integrates real-world industrial datasets, standardized evaluation metrics, and a physics-based simulation environment to support fair, reproducible, and realistic comparisons.


📄 Paper

RoboBPP: Benchmarking Robotic Online Bin Packing with Physics-based Simulation
📌 Under review at International Journal of Robotics Research (IJRR)

If you find this project useful, please consider citing our paper (see Citation).


🌐 Project Website

🔗 Project Homepage:
https://robot-bin-packing-benchmark.github.io/

The website provides:

  • Detailed benchmark descriptions
  • Dataset statistics and visualizations
  • Leaderboards across different settings
  • Documentation of evaluation metrics
  • Interactive demonstrations of the packing process

📦 Datasets

RoboBPP includes three real-world industrial datasets, covering diverse packing characteristics and challenges:

  • Repetitive Dataset – Manufacturer-oriented supply data
  • Diverse Dataset – Office supplies vendor data
  • Wood Board Dataset – Long and slender industrial materials

📥 All datasets can be downloaded from the project homepage:
https://robot-bin-packing-benchmark.github.io/download.html


🧩 Code Structure

The repository is organized into three main directories: code, test, and experiment, reflecting the full pipeline from algorithm implementation to evaluation and analysis.

code/

This directory contains the training and implementation code for all packing algorithms reproduced in this benchmark.

  • heuristics/
    Implements classical heuristic-based online bin packing methods, reproduced by us according to their original papers.

  • learning_based/
    Contains learning-based approaches, including model definitions and training pipelines, implemented following the corresponding publications.

All algorithms in this folder are our own re-implementations to ensure consistency and fair comparison across methods.


test/

This directory provides the testing and evaluation code for running all algorithms in the physics-based simulation environment.

  • Unified testing interfaces for both heuristic and learning-based methods
  • Integration with the PyBullet-based simulator
  • Scripts for executing online packing under different benchmark settings
  • Includes pre-trained models for selected learning-based methods to facilitate reproducibility

experiment/

Experiments validating the effectiveness and rationality of the proposed benchmark metrics


🚀 Getting Started

git clone https://gitlink.org.cn/wangzhoufeng/RoboBPP.git
cd RoboBPP
关于
53.3 MB
邀请码
    Gitlink(确实开源)
  • 加入我们
  • 官网邮箱:gitlink@ccf.org.cn
  • QQ群
  • QQ群
  • 公众号
  • 公众号

版权所有:中国计算机学会技术支持:开源发展技术委员会
京ICP备13000930号-9 京公网安备 11010802032778号