Skip to content

Latest commit

 

History

History
55 lines (39 loc) · 2.18 KB

README.md

File metadata and controls

55 lines (39 loc) · 2.18 KB
Raw

Homework-1

Setup and Installation

Install MuJoCo

  1. Download the MuJoCo version 2.1 binaries for Linux or OSX.
  2. Extract the downloaded mujoco210 directory into ~/.mujoco/mujoco210.
  3. Add resources/mjkey.txt in the repo into into ~/.mujoco/mujoco210.

Setup environment

To set up the project environment, Use the environment.yml file. It contains the necessary dependencies and installation instructions.

conda env create -f environment.yml
conda activate cse579a1

Install LibGLEW

sudo apt-get install libglew-dev
sudo apt-get install patchelf

Export paths variables

For these, put them in your ~/.bashrc

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/.mujoco/mujoco210/bin
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so

Compile mujoco_py (only needs to be done once)

python -c "import mujoco_py"

Running the assignment

This is the command you use to run the assignment:

python main.py --env reacher/pointmaze --train behavior_cloning/dagger/diffusion --policy gaussian/autoregressive/diffusion

For example if you wanted to run reacher with behavior cloning and gaussian policy you would run:

python main.py --env reacher --train behavior_cloning --policy gaussian

Files you need to touch:

More details in the assignment spec.

  • main.py (only for hyperparameter tuning)
    • The assignment will ask you to change certain hyperparameters in main.py like the batch_size or number of training steps.
  • DiffusionPolicy.py (for the extra credit)
    • There are three different TODO blocks to implement diffusion policies.
  • dagger.py (for your implementation of dagger)
    • There are two different TODO blocks to implement dagger.
  • bc.py (for your implementation of bc)
    • There are one different TODO blocks to implement bc.
  • utils.py (for the autoregressive model)
    • There is two different TODO blocks to implement the autoregressive model.