# RDPG-Biped

**Repository Path**: wangxiaopeng2/RDPG-Biped

## Basic Information

- **Project Name**: RDPG-Biped
- **Description**: rdpg
- **Primary Language**: Python
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-01-08
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# RDPG-Biped Code for 'Recurrent Network-based Deterministic Policy Gradient for Solving Bipedal Walking Challenge on Rugged Terrains'
https://arxiv.org/abs/1710.02896

1) Environment: Miniconda is recommended as pybox does not support pip
- python 2.7: print format might become an issue with python 3 but other than that, is fine
- numpy, scipy, matplotlib: up-to-date
- tensorflow 1.2 : higher versions are fine and TF-GPU compatible
- OpenAI gym and pybox: for gym, download the files in 'gym-files.tar.gz' and replace 'bipedal_walk.py(many other versions are provided in the tar file)' and 'time_limit.py' into the original files

2) Run default model (Our RDPG)
- learn and run: run 'gym_ddpg.py' - be sure to make proper 'checkpoint' files for both 'saved_' folders and 'gym_ddpg' folder inside 'results' directory
- record: run 'tester_r.py'
- display: run 'display.py' in 'results' directory

3) Other models
- DDPG(Feedforward network-based DPG): d3_9
- RDPG with parameter noise: r17_41_opt0
- Our RDPG with experience injection: TBA