Skip to content

Motor-learning, Representation learning, Motor Control, Deep Learning Model

Notifications You must be signed in to change notification settings

Po-Hsuan-Huang/Representation-Learning-Example

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

13 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Efficient movement representation by embedding dynamic movement premitives in deep autoencoders

Author: Po-Hsuan Huang 06/01/2018

Email : pohsuanh_at_usc.edu

This is my rotation project with Prof. Schweighofer trying to replicate the model proposed by Nutan Chen at. al. in their Efficient movement representation by embedding dynamic movement premitives in deep autoencoders (2015 IEEE-RAS). https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7363570

The model is able to reconstruct human movements by embedding dynamic movement primitives in latent space of the variational autoencoder with variational bayes fileters (Maximilian Karl et al. ICLR 2017)https://arxiv.org/abs/1605.06432.

The movement dataset comes from Master Motor Map - Framework and Toolkit for Capturing. Representing, and Reproducing Human Motion on Humanoid Robots (Oemer Terlemez et al.) https://ieeexplore.ieee.org/document/7041470/.

The project aims to use the powerful generative model to help rehabilitate patients.

What is this repository for?

These folder contains all scripts written during my rotation

The files include dynamic motor primitive example based on Matlab code from Prof. Stefan Schaal's websit

The files include stacked auto-encoders and variational auto-encoders for NMIST dataset and CIFAR10 dataset

What is not done yet ?

  • Writing the data pipeline interface from Master Motor Map for the auto-encoder, and test able to generate stationary movements

  • Integrating Dynamic Movement Primitive to the variational autoencoder, and test able to generate dynamice movments

  • Integrate deep Bayesian filters to the variational autoencoder, and test able to generate movement transitions

  • Finally, design better regularizers to learn more efficient ebeddings in latent space

How do I get set up?

Script Dependencies

  • Linxu operation system (Ubuntu)

  • Python3

  • Matplotlib

  • Numpy

  • Tensorflow-gpu

  • Spyder

Installation Guide for running Tensorflow on GPU

Detailed instruction : https://medium.com/@taylordenouden/installing-tensorflow-gpu-on-ubuntu-18-04-89a142325138

Install proper version of Nvidia GPU driver

Depending on the GPU card you have, update your GPU driver and reboot your computer.

For example, install nvidia-390 for Nvidia 1080Ti GPU.https://tecadmin.net/install-latest-nvidia-drivers-ubuntu/

You must have Nvidia GPU in order to run Tensorflow on GPU. Otherwise you can run Tensorflow on CPU.

Install with Pip

Download and install CUDA Toolkit https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

Download and install CuDNN

Then :

sudo apt-get install pip
pip install python
pip install matplotlib
pip install numpy
pip install tensorflow-gpu
pip install spyder

Install in virtual environment such as Anaconda

Virtual environemtn prevents dependencies conflicts of different softwares. For more information ask Google.

Download Anaconda and install it folloing the installation guide on https://conda.io/docs/user-guide/install/index.html

after install Anaconda, create an virtual environment 'tensorflow' in Python3 :

conda create -n tensorflow python = python3
source activate tensorflow
conda install matplotlib, spyder
conda install tensorflow-gpu

Installation Guide for running Tensorflow on CPU

With Pip

sudo apt-get install pip
pip install python
pip install matplotlib
pip install numpy
pip install tensorflow
pip install spyder

In virtual environment such as Anaconda

download Anaconda and install it folloing the installation guide on https://conda.io/docs/user-guide/install/index.html

after install Anaconda, create an virtual environment 'tensorflow' in Python3 :

conda create -n tensorflow python = python3
source activate tensorflow
conda install matplotlib, spyder
conda install tensorflow

How to run tests

Each script is independent from each other. So each of them can be executed independently by typing in your terminal :

$ python the/path/to/the/script.py

Or run script in Spyder editor by typing in your terminal :

  $ spyder

In Spyder3, open the respective file by clicking the open file icon on the toolbar in the API

However, it is recommended to run the code in Python edictor such as Spyder3 https://pythonhosted.org/spyder/ to produce images correctly.

About

Motor-learning, Representation learning, Motor Control, Deep Learning Model

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages