https://github.com/js-lee-ai/optimal-path-search_imitation-learning

Searching for optimal paths in a customized Grid-world environment using Imitation Learning; Variational Adversarial Imitation Learning [VAIL]
https://github.com/js-lee-ai/optimal-path-search_imitation-learning

imitation-learning inverse-reinforcement-learning irl pathfinding vail

Last synced: 8 months ago
JSON representation

Searching for optimal paths in a customized Grid-world environment using Imitation Learning; Variational Adversarial Imitation Learning [VAIL]

• Host: GitHub
• URL: https://github.com/js-lee-ai/optimal-path-search_imitation-learning
• Owner: js-lee-AI
• Created: 2020-03-25T18:56:46.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2022-11-23T10:27:01.000Z (over 3 years ago)
• Last Synced: 2025-04-21T17:47:59.954Z (about 1 year ago)
• Topics: imitation-learning, inverse-reinforcement-learning, irl, pathfinding, vail
• Language: Python
• Homepage:
• Size: 54.6 MB
• Stars: 20
• Watchers: 1
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Imitation Learning; Optimal Multiple Path Search Using VAIL

## How to

**The Customized Grid-World environment and actions**

![environment actions](./img/girdworld-AE.PNG)


***environment.py*** : Currently, the customized Grid-World of the 20x20 pixel window is configured.

***Expert dataset 1,2*** : Examples of configuring expert dataset with the pickle module

***expert_generator.py*** : You can use this file to create expert data.

***main.py*** : You can run this program by running ***main.py***.

## Result

### two obstacles - 10 x 10 GridWorld

You should need expert data to find approximately 50 shortest paths.

	This is a captured image executed from our old code.

**150 episode**

![two obstacle-150](./img/2obstacle-150epoch.PNG)


**500 episode**

![two obstacle-500](./img/2obstacle-500epoch.PNG)


### four obstacles - easy path

You should need expert data to find approximately 200 shortest paths.

**300 episode**

![four obstacle0-300](./img/epoch-300.png)


**500 episode**

![four obstacle0-500](./img/epoch-500.PNG)


**700 episode**

![four obstacle0-700](./img/epoch-500.PNG)


**900 episode**

![four obstacle0-900](./img/epoch-900.PNG)


**1000 episode**

![four obstacle0-1000](./img/epoch-1000.PNG)


### four obstacles - difficult path

You should need expert data to find approximately 400-500 shortest paths.

**700 episode**

![four obstacle1-700](./img/4opstacle2_epoch-700.png)


**900 episode**

![four obstacle1-900](./img/4opstacle2_epoch-900.png)


**1000 episode**

![four obstacle1-1000](./img/4opstacle2_epoch-900.png)


## Related papers

- [1] J. Ho, et al., ["Generative Adversarial Imitation Learning"](https://papers.nips.cc/paper/6391-generative-adversarial-imitation-learning.pdf), NIPS 2016.

- [2] Xue Bin Peng, et al., ["Variational Discriminator Bottleneck. Improving Imitation Learning, Inverse RL, and GANs by Constraining Information Flow"](https://arxiv.org/pdf/1810.00821.pdf), ICLR 2019.

## Reference

RL-korea : [Dongmin Lee, et al.](https://github.com/reinforcement-learning-kr/lets-do-irl)

## Author

Jungseob Lee / [ js-lee-AI](https://github.com/js-lee-AI) / omanma1928@naver.com