InterFuser: Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer

Hao Shao, Letian Wang, RuoBing Chen, Hongsheng Li, Yu Liu CoRL 2022

This repository contains code for the paper Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer. The paper has been accepted by CoRL 2022 and we will release the code soon.

This work fully processes and fuses information from multi-modal multi-view sensors for achieving comprehensive scene understanding. Besides, intermediate interpretable features are generated from our framework, which provides more semantics and is exploited to better constrain actions to be within the safe sets. Our method achieved new state-of-the-art on CARLA AD Leaderboard by June 11 2022.

Demo Video

Contents

1. Setup
2. Dataset
3. Data Generation
4. Training
5. Evaluation
6. Acknowledgements

Setup

Install anaconda

wget https://repo.anaconda.com/archive/Anaconda3-2020.11-Linux-x86_64.sh
bash Anaconda3-2020.11-Linux-x86_64.sh
source ~/.profile

Clone the repo and build the environment

git clone https://gitlab.bj.sensetime.com/shaohao/interfuser.git
cd interfuser
conda create -n interfuser python=3.7
conda activate interfuser
pip3 install -r requirements.txt
cd interfuser
python setup.py develop

Download and setup CARLA 0.9.10.1

chmod +x setup_carla.sh
./setup_carla.sh
easy_install carla/PythonAPI/carla/dist/carla-0.9.10-py3.7-linux-x86_64.egg

Note: we choose the setuptools==41 to install because this version has the feature easy_install. After installing the carla.egg you can install the lastest setuptools to avoid No module named distutils_hack.

Dataset

The data is generated with leaderboard/team_code/auto_pilot.py in 8 CARLA towns using the routes and scenarios files provided at leaderboard/data on CARLA 0.9.10.1

The collected dataset is structured as follows:

- TownX_{tiny,short,long}: corresponding to different towns and routes files
    - routes_X: contains data for an individual route
        - rgb_{front, left, right, rear}: multi-view camera images at 400x300 resolution
        - seg_{front, left, right}: corresponding segmentation images
        - depth_{front, left, right}: corresponding depth images
        - lidar: 3d point cloud in .npy format
        - birdview: topdown segmentation images required for training LBC
        - 2d_bbs_{front, left, right, rear}: 2d bounding boxes for different agents in the corresponding camera view
        - 3d_bbs: 3d bounding boxes for different agents
        - affordances: different types of affordances
        - measurements: contains ego-agent's position, velocity and other metadata
        - other_actors: contains the positions, velocities and other metadatas of surrounding vehicles and the traffic lights

Data Generation

Data Generation with multiple CARLA Servers

In addition to the dataset, we have also provided all the scripts used for generating data and these can be modified as required for different CARLA versions. The dataset is collected by a rule-based expert agent in differnet weathers and towns.

Running CARLA Servers

# start 14 carla servers: ip [localhost], port [20000 - 20026]
cd carla
CUDA_VISIBLE_DEVICES=0 ./CarlaUE4.sh --world-port=20000 -opengl &
CUDA_VISIBLE_DEVICES=1 ./CarlaUE4.sh --world-port=20002 -opengl &
...
CUDA_VISIBLE_DEVICES=7 ./CarlaUE4.sh --world-port=20026 -opengl &

Instructions for setting up docker are available here. Pull the docker image of CARLA 0.9.10.1 docker pull carlasim/carla:0.9.10.1.

Docker 18:

docker run -it --rm -p 2000-2002:2000-2002 --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=0 carlasim/carla:0.9.10.1 ./CarlaUE4.sh --world-port=2000 -opengl

Docker 19:

docker run -it --rm --net=host --gpus '"device=0"' carlasim/carla:0.9.10.1 ./CarlaUE4.sh --world-port=2000 -opengl

If the docker container doesn't start properly then add another environment variable -e SDL_AUDIODRIVER=dsp.

Run the Autopilot

Generate scripts for collecting data in batches.

cd dataset
python init_dir.py
cd ..
cd data_collection
python generate_yamls.py # You can modify fps, waypoints distribution strength ...

# If you don't need all weather, you can modify the following script
python generate_bashs.py
python generate_batch_collect.py 
cd ..

Run batch-run scripts of the town and route type that you need to collect.

bash data_collection/batch_run/run_route_routes_town01_long.sh
bash data_collection/batch_run/run_route_routes_town01_short.sh
...
bash data_collection/batch_run/run_route_routes_town07_tiny.sh

Note1: If you don't need all 14 kinds of weather in your dataset, you can modify the above code and scripts.

Note2: We also provide 7 kinds of night weather conditions in leaderboard/team_code/auto_pilot.py, you can modify the above code and scripts to collect the night dataset.

Data Generation with a single CARLA Server

With a single CARLA server, rollout the autopilot to start data generation.

carla/CarlaUE4.sh --world-port=2000 -opengl
./leaderboard/scripts/run_evaluation.sh

The expert agent used for data generation is defined in leaderboard/team_code/auto_pilot.py. Different variables which need to be set are specified in leaderboard/scripts/run_evaluation.sh.

Some Useful tools

We provide some useful python scripts in tools/data, for example:

• batch_merge_data.py: merge several fragmented data items into one item, and then interfuser/timm/data/carla_dataset.py can be modified to reduce IO time.
• batch_stat_data.py: get statistical results of collected data
• batch_preload.py: compute the waypoints of all frames in one route and save to one file
• batch_stat_blocked_data.py and batch_recollect_blocked_data.py: find and delete the frames that the ego-vehicle is blocked for a long time

Training

cd interfuser
bash scripts/train.sh

GPU_NUM=8
DATASET_ROOT='dataset/'

./distributed_train.sh $GPU_NUM $DATASET_ROOT  --dataset carla --train-towns 1 2 3 4 6 7 10  --val-towns 5 \
    --train-weathers 0 1 2 3 4 5 6 7 8 9  --val-weathers 10 11 12 13 \
    --model interfuser_baseline --sched cosine --epochs 25 --warmup-epochs 5 --lr 0.0005 --batch-size 16  -j 16 --no-prefetcher --eval-metric l1_error \
    --opt adamw --opt-eps 1e-8 --weight-decay 0.05  \
    --scale 0.9 1.1 --saver-decreasing --clip-grad 10 --freeze-num -1 \
    --with-backbone-lr --backbone-lr 0.0002 \
    --multi-view --with-lidar --multi-view-input-size 3 128 128 \
    --experiment interfuser_baseline \
    --pretrained

Evaluation

Spin up a CARLA server (described above) and run the required agent. The adequate routes and scenarios files are provided in leaderboard/data and the required variables need to be set in leaderboard/scripts/run_evaluation.sh. Update leaderboard/scripts/run_evaluation.sh to include the following code for evaluating the model on Town05 Long Benchmark.

export CARLA_ROOT=/path/to/carla/root
export ROUTES=leaderboard/data/evaluation_routes/routes_town05_long.xml
export TEAM_AGENT=leaderboard/team_code/interfuser_agent.py
export TEAM_CONFIG=leaderboard/team_code/interfuser_config.py
export CHECKPOINT_ENDPOINT=results/interfuser_result.json
export SCENARIOS=leaderboard/data/scenarios/town05_all_scenarios.json

CUDA_VISIBLE_DEVICES=0 ./leaderboard/scripts/run_evaluation.sh

Use ROUTES=leaderboard/data/42routes/42routes.xml and SCENARIOS=leaderboard/data/42routes/42scenarios.json to run CARLA 42 routes Benchmark.

Pretrain weights

We also provide examplery model weights for direct evaluation which can be downloaded at here. The model needs to be moved as model_ckpt/interfuser.pth.tar (specified by leaderboard/team_code/interfuser_config.py).

Note: The model is trained on the part of the full dataset with several towns and weathers.

Acknowledgements

This implementation is based on code from several repositories.

• Transfuser
• 2020_CARLA_challenge
• CARLA Leaderboard
• Scenario Runner

Citation

If you find our repo, dataset or paper useful, please cite us as

@article{shao2022interfuser,
 title={Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer},
 author={Hao Shao and Letian Wang and RuoBing Chen and Hongsheng Li and Yu Liu},
 journal={arXiv preprint arXiv:2207.14024},
 year={2022},
}

License

All code within this repository is under Apache License 2.0.