Skip to content

Latest commit

 

History

History
140 lines (69 loc) · 3.48 KB

carla_challenge_coil_baseline.md

File metadata and controls

140 lines (69 loc) · 3.48 KB
Raw

CARLA Challenge Track 2 Baseline - Conditional Imitation Learning

CARLA Video

CARLA Challenge Test Results

We keep an updated score for the current challenge tasks:

Challenge Basic: AVG Score 34.70

Running the Baseline

Preparation

Clone the repository:

git clone https://github.com/felipecode/coiltraine.git 
cd coiltraine

We provide a conda environment requirements file, to install and activate, just run:

conda env create -f requirements.yaml
conda activate coiltraine

Download the agent pytorch checkpoint by running the following script:

python3 tools/download_sample_models.py

The checkpoints should now be allocated already on the proper folders.

Download the latest CARLA 0.9.x version. Then, after unpacking it, define where the root folder was placed:

export CARLA_ROOT=<path_to_carla_root>

Install the latest CARLA API:

easy_install ${CARLA_ROOT}/PythonAPI/carla/dist/*-py3.5-linux-x86_64.egg

Make sure you set the PYTHONPATH PythonAPI path:

 export PYTHONPATH=${CARLA_ROOT}/PythonAPI/carla:$PYTHONPATH

Visualize the agent results

First have the latest version of CARLA executing at some terminal at 40 fps (Recommend)

sh CarlaUE4.sh Town03 -windowed -world-port=2000  -benchmark -fps=40

To run the and visualize the model run:

python3 view_model.py  -f baselines -e resnet34imnet -cp 180000 -cv 0.9

After running, you will see on the bottom corner the activations of resnet intermediate layers. You can command a destination for the agent by using the arrow keys from the keyboard.

Get the agent performance on the CARLA Challenge

Clone the scenario runner repository:

cd
git clone -b carla_challenge  https://github.com/carla-simulator/scenario_runner.git

Setup the scenario runner challenge repository by setting the path to your CARLA root folder.

cd scenario_runner
bash setup_environment --carla-root <path_to_carla_root_folder>

Export the coiltraine path to the PYTHONPATH:

cd ~/coitraine
export PYTHONPATH=`pwd`:$PYTHONPATH

Start the CARLA server on another terminal:

./CarlaUE4.sh -benchmark -fps=20 -quality-level=Epic

Execute the challenge with the conditional imitation learning baseline

 CHALLENGE_PHASE_CODENAME=dev_track_2 python3 ${ROOT_SCENARIO_RUNNER}/srunner/challenge/challenge_evaluator_routes.py \
--scenarios=${ROOT_SCENARIO_RUNNER}/srunner/challenge/all_towns_traffic_scenarios1_3_4.json \
--routes=${ROOT_SCENARIO_RUNNER}/srunner/challenge/routes_training.xml \
--debug=0 \
--agent=../coiltraine/drive/CoILBaseline.py \
--config=../coiltraine/drive/sample_agent.json

Watch the results.

Training

Define the datasets folder. This is the folder that will contain your training and validation datasets

export COIL_DATASET_PATH=<Path to where your dataset folders are>

Download the dataset:

python3 tools/get_baseline_dataset.py

You can learn how to use the framework on the following main tutorial However, you can also do a single train of the model using the basic dataset:

python3 coiltraine.py --single-process train -e resnet34imnet --folder baselines --gpus 0

To check images and train curves there is also a tensorboard log being saved at "_logs" folder on the repository root.