Understanding motion kinematics using prosthetic devices after lower limb amputation through AI-engineered model

Advanced Machine Learning Project. Catalina Botía, Isabella Ramos, Daniela Tamayo.

Universidad de los Andes, 2020

Getting Started

Clone this repository. Create a new environment following these commands:

$ conda create -n opensim -c kidzik -c conda-forge opensim python=3.6.2
$ conda activate opensim
$ conda install pytorch torchvision torchaudio cudatoolkit=10.1 -c pytorch
$ pip install git+https://github.com/stanfordnmbl/osim-rl.git@ver2.1
$ pip install matplotlib

Environment

The environment used is taken from the NeurIPS 2018 Challenge: Learning to Run which is modified with a prosthetic leg for the purpose of this project.

Running the code

Make sure to check the arguments.py file. In the arguments graphs_folder and checkpoint_dir choose a name for the folder to save you experiment.

To train

$ CUDA_VISIBLE_DEVICES=0 python main.py --render_environment=False --mode_test=False 

To evaluate

$ python Test.py

The final trained model for best and last rewards can be found in this link or in BCV002 /media/user_home0/dtamayo/AMLproject/saved_model_final. Download it and make sure to have it saved on a folder with the same name as the checkpoint_dir parameter in the arguments.py file. Run on MobaXTerm to visualize the skeleton!

The final presentation is also attached on the respository (though it doesn't have animations) as well as the CVPR article.