cd $HOME
git clone https://github.com/avillaflor/P2DBM.git
cd P2DBM
conda env create -f environment.yml
conda activate p2dbm
pip install -e .
cd $HOME
git clone https://github.com/autonomousvision/plant.git
Install CARLA v0.9.10 (https://carla.org/2020/09/25/release-0.9.10/) for which the binaries are available here: (https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/CARLA_0.9.10.tar.gz)
mkdir $HOME/carla910
cd $HOME/carla910
wget "https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/CARLA_0.9.10.tar.gz"
tar -xvzf CARLA_0.9.10.tar.gz
wget "https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/AdditionalMaps_0.9.10.tar.gz"
tar -xvzf AdditionalMaps_0.9.10.tar.gz
rm CARLA_0.9.10.tar.gz
rm AdditionalMaps_0.9.10.tar.gz
Install CARLA v0.9.11 (https://carla.org/2020/12/22/release-0.9.11/) for which the binaries are available here: (https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/CARLA_0.9.11.tar.gz)
mkdir $HOME/carla911
cd $HOME/carla911
wget "https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/CARLA_0.9.11.tar.gz"
tar -xvzf CARLA_0.9.11.tar.gz
wget "https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/AdditionalMaps_0.9.11.tar.gz"
tar -xvzf AdditionalMaps_0.9.11.tar.gz
rm CARLA_0.9.11.tar.gz
rm AdditionalMaps_0.9.11.tar.gz
Add the following to your .bashrc:
export CARLA_9_10_PATH=$HOME/carla910
export CARLA_9_10_PYTHONPATH=$CARLA_9_10_PATH/PythonAPI/carla/dist/carla-0.9.10-py3.7-linux-x86_64.egg
export CARLA_9_11_PATH=$HOME/carla911
export CARLA_9_11_PYTHONPATH=$CARLA_9_11_PATH/PythonAPI/carla/dist/carla-0.9.11-py3.7-linux-x86_64.egg
./scripts/merge_scenarios/collect_data.sh
python scripts/merge_scenarios/create_h5_dataset.py
python scripts/merge_scenarios/ours/train.py gpu=[0]
./scripts/merge_scenarios/ours/closed_planner_eval.sh 'MODEL_CKPT_LOCATION'
./scripts/merge_scenarios/ours/eval.sh 'MODEL_CKPT_LOCATION'
./scripts/merge_scenarios/ours/open_planner_eval.sh 'MODEL_CKPT_LOCATION'
Fill in MODEL_CKPT_LOCATION with location of .ckpt file for model you want to evaluate.
python scripts/merge_scenarios/vae/vae_train.py gpu=[0]
./scripts/merge_scenarios/vae/vae_closed_planner_eval.sh 'MODEL_CKPT_LOCATION'
python scripts/merge_scenarios/plant/train.py gpu=[0]
./scripts/merge_scenarios/plant/eval.sh 'MODEL_CKPT_LOCATION'
Add user config at scripts/leaderboard/config/user/$USER.yaml and fill in working_dir with path of P2DBM directory, carla_path with path to CARLA version 0.9.10, and plant_dir with path to PlanT directory.
For running longest6 experiments, you need a separate CARLA instance running when collecting data or running evaluations.
cd $HOME/carla910
SDL_VIDEODRIVER=offscreen SDL_HINT_CUDA_DEVICE=0 ./CarlaUE4.sh --world-port=2000 -traffic-port=8000 -opengl
./scripts/leaderboard/datagen.sh 2000 8000 1
./scripts/leaderboard/datagen.sh 2000 8000 2
./scripts/leaderboard/datagen.sh 2000 8000 3
./scripts/leaderboard/datagen_valid.sh 2000 8000
python scripts/leaderboard/ours/create_h5_dataset.py
python scripts/leaderboard/ours/train.py gpu=[0]
python scripts/leaderboard/run_evaluation.py user=$USER experiments=closed_planner eval=longest6 port=2000 trafficManagerPort=8000 CUDA_VISIBLE_DEVICES=0 experiments.model_checkpoint=MODEL_CKPT_LOCATION
python scripts/leaderboard/run_evaluation.py user=$USER experiments=IL eval=longest6 port=2000 trafficManagerPort=8000 CUDA_VISIBLE_DEVICES=0 experiments.model_checkpoint=MODEL_CKPT_LOCATION
Fill in MODEL_CKPT_LOCATION with location of .ckpt file for model you want to evaluate.
Code for running leaderboard and PlanT comparisons comes from PlanT repo.
Transformer code initially adapted from Autobots repo.
Testing environments are based on the CARLA simulator and Leaderboard v1.0 benchmark.