Re4MPC: Reactive Nonlinear MPC for Multi-model Motion Planning via Deep Reinforcement Learning

Pick-and-Place

Manual Target

Training

Testing

1. Citation

@inproceedings{akmandor2025re4mpc,
  title={Re4MPC: Reactive Nonlinear MPC for Multi-model Motion Planning via Deep Reinforcement Learning},
  author={Akmandor, Ne{\c{s}}et {\"U}nver and Prajapati, Sarvesh and Zolotas, Mark and Padir, Ta{\c{s}}kin},
  booktitle={2025 IEEE International Conference on Automation Science and Engineering (CASE)},
  year={2025},
  organization={IEEE}
}

2. Related Papers

• Akmandor, N.{"U}., Prajapati, S., Zolotas, M. and Padır, T., 2025. Re4MPC: Reactive Nonlinear MPC for Multi-model Motion Planning via Deep Reinforcement Learning. arXiv preprint arXiv:2506.08344.

3. Videos

• CASE 2025 - supplementary video
• CASE 2025 - presentation

4. Installation

4.1 Clone libccd into any directory (e.g. Home) and install libccd library from its source:

git clone https://github.com/danfis/libccd.git
cd libccd
mkdir build && cd build
cmake -G "Unix Makefiles" -DENABLE_DOUBLE_PRECISION=ON ..
sudo make -j4
sudo make install

4.2 Clone fcl into any directory (e.g. Home) and install fcl library from its source:

git clone https://github.com/flexible-collision-library/fcl.git
cd fcl
mkdir build && cd build
cmake ..
sudo make -j4
sudo make install

4.3 Clone the mobiman repository into the src folder of catkin workspace and install all required ROS packages:

git clone [email protected]:RIVeR-Lab/mobiman.git
cd mobiman
./install_mobiman.sh

4.4 Install ROS dependencies:

rosdep install --from-paths src --ignore-src -r -y

4.5 Install iGibson simulator and its library by following the instructions in igibson README.

5. Run

5.1 Manual mode

5.1 Set configurations in mobiman_framework.launch

Set parameter "config_mobiman_framework":

• iGibson example: "config_mobiman_framework_igibson_manual"

5.2. In seperate terminals in ROS workspace:

5.2.1 Launch mobiman framework:

# Do not forget to source: source devel/setup.bash
roslaunch mobiman_simulation mobiman_framework.launch 2> >(grep -v TF_REPEATED_DATA buffer_core)

5.2.2 Run the (python) simulation script:

# Do not forget to source: source devel/setup.bash
python src/igibson-ros/scripts/mobiman_jackalJaco.py

Note: grep command added to avoid stream of warnings as depicted in here.

5.2 DRL mode

5.2.1 Set configurations in mobiman_framework.launch

5.2.1.1 Set parameter "config_mobiman_framework":

    - iGibson example: "config_mobiman_framework_igibson_drl"

5.2.1.2 Set drl configurations in config_mobiman_drl_pick

5.2.2 In seperate terminals in ROS workspace:

5.2.2.1 Launch mobiman framework:

# Do not forget to source: source devel/setup.bash
roslaunch mobiman_simulation mobiman_framework.launch 2> >(grep -v TF_REPEATED_DATA buffer_core)

5.2.2.3 Run the (python) training/testing script:

Training:

# Do not forget to source: source devel/setup.bash
python src/igibson-ros/scripts/drl_training_sb3_mobiman_jackalJaco.py

Testing:

# Do not forget to source: source devel/setup.bash
python src/igibson-ros/scripts/drl_testing_sb3_mobiman_jackalJaco.py

6. Dataset

Trained models to reach static object on a conveyor belt

7. Code Contributors

Neset Unver Akmandor | [email protected]
Sarvesh Prajapati | [email protected]

8. Credentials

mobiman was developed at the RIVeR Lab, Northeastern University.