NNMPC

Intention-based NNMPC controller for CARLA simulator. NNMPC is a mathematical control framework based on Model Predictive Control (MPC) encompassing a state-of-the-art Recurrent Neural network (RNN) architecture. (SCRIPTS TO BE PUBLISHED)

S. Bae, A. Nakhaei, C. Choi, D. Saxena, K. Fujimura, and S. Moura, Cooperation-Aware Lane Change Maneuver in Dense Traffic based on Model Predictive Control with Recurrent Neural Network, preprint.

@article{bae2019cooperation,
  title={Cooperation-Aware Lane Change Maneuver in Dense Traffic based on Model Predictive Control with Recurrent Neural Network},
  author={Bae, Sangjae and Saxena, Dhruv and Nakhaei, Alireza and Choi, Chiho and Fujimura, Kikuo and Moura, Scott},
  journal={arXiv preprint arXiv:1909.05665},
  year={2019}
}

Prerequisites

Setup

Create a virtual environment

conda create -n carla-env python=2.7
conda activate carla-env

Install carla-lane-change-setup

git clone https://github.com/honda-research-institute/carla-lane-change-setup.git
cd carla-lane-change-setup
pip install -r requirements.txt
./setup_workspace.sh
cd catkin_ws
catkin_make

Install dependencies of SGAN

pip install -r requirements.txt   # Install dependencies

Install rospy [ubuntu]

sudo apt-get update -y
sudo apt-get install -y python-rospy

Project structure

nnmpc_source: contains NNMPC files
- nnmpc.py: MPC with SGAN
- lane.py: Lane class
- vehicle.py: Vehicle class
- lane_keep_opt_planner.py: MPC-based smooth trajectory planner
- path_optimizer.py: solve MPC for smooth trajectory
- extended_kalman_filter.py: extended Kalman Filter
- utils.py: helper functions
- frenet_utils.py: helper functions for frenet coordinates
sgan_source: contains source code of SGAN and trained model
- predictor.py: contains the Predictor class, which loads trained SGAN model
config.py: hyper parameters. Descriptions are found here
init_ros_node.py: run NNMPC as a ROS node

Run NNMPC with Carla + scenario runner

Step 1: Start Carla Simulator (carla documentation: http://carla.readthedocs.io/en/latest/).

source ~/.bashrc
conda activate carla-env
source ~/carla-lane-change-setup/catkin_ws/devel/setup.bash
sudo nvidia-docker run -p 2000-2001:2000-2001 -it --rm carlasim/carla:0.9.6 ./CarlaUE4.sh

Step 2: Launch ROS Bridge. In a new terminal

source ~/.bashrc
conda activate carla-env
source ~/carla-lane-change-setup/catkin_ws/devel/setup.bash
roslaunch demo_entrance carla_ros_bridge_with_ego_vehicle.launch

More information about the Carla ros bridge can be found on the Official Repo.

Step 3: Launch NNMPC. In a new terminal

source ~/.bashrc
conda activate carla-env
source ~/carla-lane-change-setup/catkin_ws/devel/setup.bash
python ~/NNMPC_CARLA/init_ros_node.py

Step 4: Start lane change scenario under scenario_runner to test your lane change algorithm.

use the following parameters: --dense: true or false (dense traffic vs spars traffice) --cooperative: true or false (more cooperative drivers vs less cooperative drivers) --repetitions: n

git clone git@github.com:honda-research-institute/scenario_runner.git
cd scenario_runner
source ~/.bashrc
conda activate carla-env
source ~/carla-lane-change-setup/catkin_ws/devel/setup.bash
python ~/scenario_runner/scenario_runner.py --scenario IntentionAwareLaneChange_3 --cooperative false --dense true --repetition 30 --waitForEgo

The above commands are tested with Carla version 0.9.6 with Python 2.7 kernel on Ubuntu 16.04.

Author

Developed by Sangjae Bae as part of his internship at Honda Research Institute, USA.

Acknowledgements

Supervisor: Alireza Nakhaei, David Isele
Collaborator: Peng Xu, Alexandre Miranda Anon, Kikuo Fujimura

HutchLiu/NNMPC