tum-adlr-ss22-07

Install using Docker

1. Install Docker following the instructions on the link and nvidia-docker (for gpu support).

2. Clone this repo

3. Build Docker Container

   docker build . -t reinforcement_planning

Run (Needs nvidia-docker and the right Nvidia GPU drivers)

source run_docker.sh 

Training

python3 train.py --config-file configs/sac.yaml # You can replace sac.yaml by ddpg.yaml or ppo.yaml

You can modify the .yaml config file to experiment with static/dynamic obstacles, number of obstacles, etc.

Testing

python3 eval.py --experiment sac/00 # You can replace sac by ddpg or ppo

The number 00 can be replaced by the id of the folder where the experiment is saved. (For every training, a new folder is created with an increasing number)

Testing differential model:

python3 eval.py --experiment sac/00_diff # You can replace sac by ddpg or ppo

Watch training curves in tensorboard

tensorboard --logdir ./experiments

Installing In host machine (with conda)

1. Create conda environment

   conda create -n "reinforcement_planning" python=3.8.10

2. Activate conda environmnet

   conda activate reinforcement_planning

3. Install dependencies

   python -m pip install -r requirements.txt

4. Install pytorch

   pip install --no-cache-dir torch==1.10.0+cu113 torchvision==0.11.0+cu113 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.html

5. Install further packages

   python -m pip install -e nav2D-envs/
   python -m pip install -e rlkit/