Coopernaut

COOPERNAUT: End-to-End Driving with Cooperative Perception for Networked Vehicles
Jiaxun Cui*, Hang Qiu*, Dian Chen, Peter Stone, Yuke Zhu
CVPR 2022

Paper | Website | Bibtex | Dataset

Introduction

Optical sensors and learning algorithms for autonomous vehicles have dramatically advanced in the past few years. Nonetheless, the reliability of today's autonomous vehicles is hindered by the limited line-of-sight sensing capability and the brittleness of data-driven methods in handling extreme situations. With recent developments of telecommunication technologies, cooperative perception with vehicle-to-vehicle communications has become a promising paradigm to enhance autonomous driving in dangerous or emergency situations.

We introduce COOPERNAUT, an end-to-end learning model that uses cross-vehicle perception for vision-based cooperative driving. Our model encodes LiDAR information into compact point-based representations that can be transmitted as messages between vehicles via realistic wireless channels. To evaluate our model, we develop AutoCastSim, a network-augmented driving simulation framework with example accident-prone scenarios. Our experiments on AutoCastSim suggest that our cooperative perception driving models lead to a 40% improvement in average success rate over egocentric driving models in these challenging driving situations and a 5 times smaller bandwidth requirement than prior work V2VNet.

Installation

First, clone the Coopernaut repo with AutoCastSim submodule.

git clone --recursive [email protected]:UT-Austin-RPL/Coopernaut.git

We provide a quickstart installation script in scripts/install.sh. In the root folder of this repo, run

source scripts/install.sh #Requires CUDA11.0

or

source scripts/install-cu113.sh #Requires CUDA11.3

If you encounter errors or would like to follow individual steps, please refer to INSTALL.md for more details.

If the the installation of mosquitto failed and you have sudo priority, please try the following command

sudo apt-get install mosquitto libopenblas-dev

Quick Start

We provide a quick example to create evaluation trajectories of a trained Coopernaut model under the scenario 6:Overtaking, with two parallel threads(Make sure your GPU has a memory larger than 6GB, otherwise change the CARLA_WORKERS in the scripts/quick_run.sh to 1). You can check the saved trajectories under Coopernaut/result directory.

conda activate autocast
./scripts/quick_run.sh

Usage

Our scripts are separated based on models. For example, all data collection, DAgger, training and evaluation of Coopernaut codes are contained in the script run_cooperative_point_transformer.sh.

bash run_cooperative_point_transformer.sh

To checkout a specific method, please replace the script with the method's corresponding pipeline script

The below instructions apply to all scripts, and all the modification should be in the scripts.

First, in the scripts, specify the root path you wish to store behavior cloning data from: TrainValFolder, and the path to store DAgger data from: DATAFOLDER. By default they are

TrainValFolder=./data/AutoCast_${SCEN}
DATAFOLDER=./data/AutoCast_${SCEN}_Small

for each scene.

Data Collection

Download Dataset

Before training, you may download the behavior cloning data here, or collect your own data by running the scripts provided. In the first prompt, select data-train, data-val or data-expert and select the scene number:

Behavior cloning training

Run this after you have both training and validation data collected in the previous step. In the first prompt, select bc.

DAgger training

Specify the behavior cloning model checkpoint path trained in the previous stage CHECKPOINT, and then when running script, select dagger in the prompt.

Evaluation

Download Pretrained Models

Specify RUN= to be the path of your DAgger model path. Then, select eval as the prompt. Then get the metrics by running

python scripts/result_analysis.py --root ${PATH_TO_YOUR_OUTPUT_DIR}
python scripts/compare_trajectory.py --eval ${PATH_TO_YOUR_OUTPUT_DIR} --expert ${PATH_TO_EXPERT}

Citation

@inproceedings{cui2022coopernaut,
    title = {Coopernaut: End-to-End Driving with Cooperative Perception for Networked Vehicles},
    author = {Jiaxun Cui and Hang Qiu and Dian Chen and Peter Stone and Yuke Zhu},
    booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    year = {2022}
}