I ported both Nullspace MPC and MPPI-H from ROS1 Noetic to ROS2 Jazzy with Gazebo 8.10.0 on Ubuntu 24.04 (native, without Docker), while preserving the original RViz plugin style. Both algorithms have been verified with navigation and demo commands.

I ran the experiment 3 times with a Maze Map. Each episode has 8 sequential goals. The table below compares the ROS2 Jazzy implementation against the original ROS1 Noetic implementation. Full data is available in maze20260405.csv.

Key improvements in the ROS2 Jazzy port:

• Nullspace MPC now reaches all goals (previously failed in the ROS1 Noetic maze environment).
• Lower mean state cost for both algorithms compared to their ROS1 counterparts.
• ~5x faster computation for Nullspace MPC (138 ms vs. 739 ms).

In conclusion, this repository successfully ports Nullspace MPC and MPPI-H to ROS2 Jazzy and Gazebo 8.10.0. Future work will focus on testing in more practical environments.

Nullspace MPC is a novel multi-objective control framework for explicitly handling task priorities. It demonstrates agile and smooth navigation through a narrow environment with a swerve drive vehicle.

Note: This is the recommended setup for optimal performance.

This package supports two operation modes:

1. Manual Goal Mode — you set a 2D Nav Goal in RViz and the robot navigates to it.
2. Demo (Multi-Goal) Mode — the robot automatically visits a sequence of goals defined in an agenda file.

Set a 2D Nav Goal in RViz, and the robot will navigate to the goal.

• Nullspace MPC (Proposed)

  cd <path-to-your-workspace>/nullspace_mpc
  make navigation_nullspace_mpc

  Try this setting for reduced computational load.

  cd <path-to-your-workspace>/nullspace_mpc
  make navigation_nullspace_mpc_lite

• MPPI (Baseline)

  cd <path-to-your-workspace>/nullspace_mpc
  make navigation_mppi

Runs an evaluation script that automatically sends multiple goals in sequence (defined in data/eval_demo/agenda.yaml) and logs results to ./result/.

• Nullspace MPC (Proposed)

  cd <path-to-your-workspace>/nullspace_mpc
  make eval_demo_nullspace_mpc

• MPPI (Baseline)

  cd <path-to-your-workspace>/nullspace_mpc
  make eval_demo_mppi

If you use this software in your research, please consider citing the following papers.

Nullspace MPC is a core component of my Ph.D. dissertation, which was made publicly available on October 1, 2025. You can access the full paper via the project website.

  @phdthesis{mizuho2025phd,
    author    = {Mizuho Aoki},
    title     = {{Nonlinear Model Predictive Control for Autonomous Vehicles: Enhancement via Simplified Physics-Aware Prediction and Decomposed Optimization}},
    school    = {Nagoya University},
    year      = {2025},
    type      = {Ph.D. Dissertation},
    language  = {English},
    url       = {https://mizuhoaoki.github.io/projects/nullspace_mpc}
  }

The baseline MPPI controller is based on the following work and is available in a dedicated repository: mppi_swerve_drive_ros

@inproceedings{mizuho2024iros,
  author={Aoki, Mizuho and Honda, Kohei and Okuda, Hiroyuki and Suzuki, Tatsuya},
  booktitle={2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, 
  title={{Switching Sampling Space of Model Predictive Path-Integral Controller to Balance Efficiency and Safety in 4WIDS Vehicle Navigation}}, 
  year={2024},
  pages={3196-3203},
  doi={10.1109/IROS58592.2024.10802359}
}

This project is licensed under the MIT License, with the exception of its core QP solving capability, which is provided by QpSolverCollection, located in the src/third_party directory and licensed under the BSD 2-Clause License.

Special thanks to the authors of QpSolverCollection for their excellent work.