Clone the RL-ViGen repo:
git clone https://github.com/gemcollector/RL-ViGen.git
cd RL-ViGen/
Create a conda environment:
conda create -n rl-vigen python=3.8
Run the installation script:
bash setup/install_rlvigen.sh
In addition, the following resources are required:
-
CARLA: We apply CARLA 0.9.10 which is a stable version in RL-ViGen. The CARLA 0.9.10 should be downloaded first, and place it to the
./third_partyfolder:wget https://carla-releases.s3.eu-west-3.amazonaws.com/Linux/CARLA_0.9.10.tar.gz -
Robosuite: we have employed the
1.4.0version of Robosuite, concurrently utilizing mujoco version2.3.0as the underlying simulator engine. We have incorporated all the relevant components associated with Robosuite in the first creating conda step. -
DM-Control: Our DM-Control also contains mujoco_menagerie as the basic component. We have incorporated all the relevant components associated with DM-Control in the first creating conda step.
To use the Habitat, you need to set up a separate conda environment:
conda create -n vigen-habitat python=3.8 cmake=3.14.0 -y
bash setup/install_vigen-habitat.sh
-
We are using Gibson scene datasets for our experiment. You can find instructions for downloading the dataset here.
-
Next we need the episode dataset for the experiments. You can get the training and validation dataset from here and place it in the ./data folder under the path :
data/datasets/pointnav/gibson/v1/.
We provide a script to install all the dependencies and resources required for RL-ViGen. You can run the following command to install:
bash setup/install_all.sh
If you are employing this method of installation, please modify all the train and eval scripts, ensuring that the parameter task is set to task@_global_. Here is an example:
CUDA_VISIBLE_DEVICES=0 python train.py \
env=${env} \
task@_global_=${task_name}
docker pull cititude/rl-vigen:1.0