/VEXAI-Sim

Primary LanguageC#

VEXAI Simulation and Training

Check out our technical report here!

Simulation

  1. Download Repo
  2. Launch repo in latest unity version

Training

  1. Install anaconda python virutal enviroment manager - conda on package managers, lookup anaconda online for windows download
  2. Navigate to where you downloaded the simulation
  3. Run command inside terminal in linux or anaconda powershell prompt on windows conda create --name name python=3.6
  4. Run command conda activate name
  5. Install required python packages to run the Unity ml-agents package a. Run python -m pip install mlagents==0.26.0 inside your virtual environment, if on windows PyTorch will have to be installed separetly and you may have to run command sudo if errors https://github.com/Unity-Technologies/ml-agents/blob/main/docs/Installation.md
  6. Press the play button in the unity editor
  7. Run mlagents-learn mlagents-config.yaml to train on our hyperparameters Edit the mlagents-config.yaml file for tuning your own hyperparameters

Tensorboard logging

  1. Open a separate terminal from the one that is training the model
  2. Navigate to the directory where the simulation is, activate the virtual enviroment, and run tensorboard --logdir /results
  3. In a browser enter "http://localhost:6006/" for your tensorboard stats