Stochastic-Guidance

Learning with Stochastic Guidance for Navigation

By Linhai Xie, Yishu Miao, Sen Wang, Phil Blunsom, Zhihua Wang, Changhao Chen, Niki trigoni, Andrew Markham.

The tensorflow implmentation for the paper: Learning with Stochastic Guidance for Navigation

Contents

0. Introduction
1. Prerequisite
2. Instruction
3. Citation

Introduction

In this project we proposed a stochastic switching machanism which is an improved version of our ICRA paper.

The stochastic switching network is implemented with a fully connected network and trained with REINFORCE algorithm.

For details please see the paper

The implementation of DDPG is based on Emami's work.

Prerequisites

Tensorflow > 1.1

ROS Kinetic

ros stage

matplotlib

cv2

Instruction

roscore

rosrun stage_ros stageros PATH TO THE FOLDER/AsDDPG/worlds/Obstacles.world

python DDPG_stochastic.py

You can download the prioritized replay buffer here or run the above training command once. It takes random actions or selections of controllers to initialize the replay buffer and create a pickle file.

Citation

If you use this method in your research, please cite:

@article{xie2018learning,
	title={Learning with Stochastic Guidance for Navigation},
	author={Xie, Linhai and Miao, Yishu and Wang, Sen and Blunsom, Phil and Wang, Zhihua and Chen, Changhao and Markham, Andrew and Trigoni, Niki},
	journal={arXiv preprint arXiv:1811.10756},
	year={2018}}