Full-Body Torque-Level Nonlinear Model Predictive Control for Aerial Manipulation

This repository contains the needed code to reproduce the experiments presented in the paper Full-Body Torque-Level Nonlinear Model Predictive Control for Aerial Manipulation.

Software dependencies

EagleMPC

• Install the EagleMPC library at this tag and its dependencies.

⚠️ Crocoddyl version ⚠️ To get the same results as in the paper, you should checkout the Crocoddyl repository to this tag.

EagleMPC-ROS

The experiments to test the MPC controllers have been run in a simulated environment involving Gazebo and ROS Noetic. Make sure both are installed intro your computer.

Then, you need to clone and download the ROS packages in EagleMPC-ROS. Follow the installation instructions from the repository.

Experiments

To run the different experiments you first need to clone this repository into your computer.

cd <choose-your-path>
git clone https://github.com/PepMS/fbtl_nmpc_experiments.git

Trajectory optimization

To run a trajectory optimization experiment you need to do 2 steps:

1. Modify the trajectory .yaml file with your paths. For example, open the hexacopter370_flying_arm_3_eagle_catch.yaml and change the urdf and the follow fields. Substitute the text between <> with the actual paths to the respective libraries.

trajectory:
  robot:
    name: "hexacopter_370_flying_arm_3"
    urdf: "<path-to-example-robot-data>/example-robot-data/robots/hexacopter370_description/urdf/hexacopter370_flying_arm_3.urdf"
    follow: "<path-to-ros-ws>/src/eagle_mpc_ros/eagle_mpc_yaml/multicopter/hexacopter370.yaml"

2. Execute the selected Python script (with the display option in case that you want to visualize this in the Gepetto-Viewer). For example, for the Eagle's Catch case:

cd <choose-your-path>/fbtl_nmpc_experiments/trajectory-optimization
python3 eagle_catch.py display

nonLinear Model Predictive Control

To run an nMPC experiment. For example, the 4-Displacement experiment.

1. Modify the file mpc/displacement/hexacopter370_flying_arm_3_mpc.yaml, which is placed inside the same folder as the Python script. Then, substitute the text between <> with the actual paths to the respective libraries.

   trajectory:
     robot:
       name: "hexacopter_370_flying_arm_3"
       urdf: "<path-to-example-robot-data>/example-robot-data/robots/hexacopter370_description/urdf/hexacopter370_flying_arm_3.urdf"
       follow: "<path-to-ros-ws>/src/eagle_mpc_ros/eagle_mpc_yaml/multicopter/hexacopter370.yaml"

2. Analogously, modify the yaml file associated to the trajectory, either 4-displacement (hexacopter370_flying_arm_3_displacement.yaml for the Carrot and Rail case and hexacopter370_flying_arm_3_displacement_w.yaml for the Weighted controller) or the eagle_catch (hexacopter370_flying_arm_3_eagle_catch_nc.yaml). These are placed inside the eagle_mpc_yaml ROS package.
3. Open the script of the experiment you want to run. For example mpc/displacement/displacement.py
4. Set the variables in the section # -----VARIABLES----- to match your settings.
   1. mpcController allows you to select among the nMPC controllers (Weighted, Rail and Carrot)
   2. controller_settings_path is the path of the .yaml file contained in the same folder of the Python script
   3. controller_settings_destination is the location of the eagle_mpc_yaml ROS package
   4. For the disturbance case, you can also set the properties of the simulated disturbance