This is a project for solving bin packing problems, which use methods like heuristic searching and reinforcement learning.

The demo using heuristic searching method has been completed, you can run test.py to see how it works. The part of reinforcement learning hasn't been completed and I can't guarantee when can I finish it. (Maybe never)

Below are explanations to some files:

1. utils.py provides basic classes that many other files may use, including Position, Attitude, Transform and other classes concerning the status of a single object.

2. object.py provides definition of class Geometry and Item, Geometry mainly contains geometric transformation and heightmap calculation. Item wraps up Geometry , makes it more accessible to upper classes.

3. container.py defines class Container, it provides manipulations of searching possible positions and adding an item to the container.

4. pack.py is the top-level class, it defines how to solve a bin packing problem, it provides functions about loading a sequence of objects, automatically adding them to a container, and clearing the container.

5. show.py defines class Display, it visualizes the procedure of bin packing decision. You only need to pass an object of type Geometry to its function show3d(), then it will show you how this geometry looks like.

6. test.py provides a demo that use heuristic method to solve bin packing problem. If it runs successfully, you will see the window as below.

   

7. train.py, env.py and common/, DQN/ are used for the reinforcement learning method, but it hasn't complete, so these files and folders remain unused.

I have exported my anaconda environment to the file environment.yaml, use conda env create -f environment.yaml to copy my environment to your computer.

This idea of this project comes from the paper Stable bin packing of non-convex 3D objects with a robot manipulator, Fan Wang and Kris Hauser. (arXiv:1812.04093v1 [cs.RO] 10 Dec 2018) , and this paper is attached in the literature folder, you can check it for detailed information.

Yang Ding

Department of Automation, Tsinghua University

Email: [email protected]

Web: https://yang-d19.github.io/