Contingency-MPPI: Contingency Constrained Planning with MPPI within MPPI

Accepted to L4DC 2025

Dependencies

Our code has been tested on Ubuntu 22.04. If you wish to run our code without use of Docker, please install the following dependencies:

sudo apt-get install libcdd-dev libblas3 libblas-dev liblapack3 liblapack-dev gfortran ccache libasio-dev 

If you also wish to run our code on hardware or Gazebo via ROS2-Humble:

sudo apt-get install ros-humble-gazebo-ros-pkgs ros-humble-navigation2

To use with SNOPT, request a trial license here. After receving the license file and libsnopt zip file, unzip the file, and add both to your environment variables (it may be convenient to add these lines to a ./bashrc file):

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:{path to unzipped libsnopt folder}
export SNOPT_LICENSE={path to license file}

Setup (No ROS)

This is the setup to replicate our simulation results in our paper:

1. Clone Repository

git clone https://github.com/neu-autonomy/Contingency-MPPI
cd Contingency-MPPI

2. Install requirements

pip3 install -r requirements.txt
# Optionally, if you wish to run with jax-cuda (recommended), run
# pip install -U "jax[cuda12]"

3. Install CMPPI

pip install -e .

Setup (with ROS)

This is the setup to run our code within a gazebo environment (and with minimal changes, on hardware). As our tests are with the Agile-X Scout-Mini, we will install the tools needed to work with the Scout-Mini, as well as the Ros2-Nav2 stack and Gazebo

1. Clone Repository into a workspace

mkdir -p ws/src && cd ws/src
git clone https://github.com/neu-autonomy/Contingency-MPPI

2. Clone Scout/UGV packages

git clone https://github.com/westonrobot/ugv_sdk.git
git clone https://github.com/leonardjmihs/scout_ros2

3. Install the Nav2 stack/Gazebo

sudo apt-get install ros-humble-gazebo-ros-pkgs ros-humble-navigation2

4. Build

cd ..
colcon build

Running Random Simulations

To recreate the sim results in our paper, run nmppi_global_random.py in 'branch_mppi/jax_mppi/'

cd branch_mppi/jax_mppi
python nmppi_global_random.py --solver snopt

This will generate a random scenario and run each variant of our algorithms as well as the baseline algorithms, and save the results in a folder titled "sim_results*/""

Running Gazebo Simulation

To run a sample of our algorithm on Gazebo, first open 3 terminals, and run the following commands (don't forget to source the workspace):

1. ros2 launch scout_ugv_sim empty_world.launch.py
2. ros2 launch branch_mppi nav2_costmap.launch.py
3. ros2 launch branch_mppi branch_mppi_node_sim.launch.py

You can visualize via rviz2: we provide a helpful rviz param file in config/branch_mppi_view.rviz

Citation

When using CMPPI, please cite the following paper (pdf, video)

@misc{jung2024contingencyconstrainedplanningmppi,
      title={Contingency Constrained Planning with MPPI within MPPI}, 
      author={Leonard Jung and Alexander Estornell and Michael Everett},
      year={2024},
      eprint={2412.09777},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2412.09777}, 
}