Perception-driven Formation Control of Airships for project AirCap using nonlinear model predictive control (MPC)

A preprint of our method is available on arXiv: https://arxiv.org/abs/2209.13040

This video is giving an overview of our approach:

Usage

1. Prepare a system matching the pre-requirements
2. Download this sourcecode repository into your home folder
3. Install the compilation requirements and compile code
4. Run the simulation demo or the MPC evaluation demos

Pre Requirements

This code has been tested with Ubuntu Linux 20.04 LTS (Focal Fossa) and ROS Noetic Ninjemys only. The Neural Network detector used for person detection requires an NVIDIA GPU with current Cuda capable drivers installed.

Note:
Without GPU, the SSD detector will not run, and the airships will not "see" the person used as a subject in the demo. They will then assume the subject at the center of the map and circle that stationary point instead.

Download

You can download the code with the command git clone https://github.com/robot-perception-group/Airship-MPC

Compile

Please run the script ./install_and_compile.sh it will install all necessary requirements including ROS Noetic Ninjemys and compile the code.

Execute

• The Gazebo simulation with wind and person tracking can be started with cd experiments/sim; ./airship_sim.sh 3
• The MPC can be tested standalone using it's own motion model for simulation with cd src; python3 blimps_costtest.py
• The solver generated trajectories and plots from our paper can be recreated with the scripts found in the subfolders in folder experiments/trajectory

Modify

• A simulation with a stationary person can be started with
  cd experiments/sim; ./airship_sim.sh 3 100 test arena_BLIMP_stat_target
• Constraints and properties of the MPC can be modified in src/blimp_nmpc_wrapper_node/nodes/formation_config.py including the number of airships in the formation. Whenever this is changed, a recompilation is necessary.
• submodules/AirCap/packages/3rdparty/airship_simulation/deflate_blimp.sh can be used to alter the airship buoyancy and rigidity
• Wind and other environmental parameters affecting the airships can be modified by editing submodules/AirCap/packages/3rdparty/airship_simulation/blimp_description/urdf/description_plugin.xacro. This folder also includes the physical model description of the airship and it's aerodynamic properties in Gazebo URDF format.
• The scripts in submodules/AirCap/scripts/simulation/ can be used to run various other experiments, including standard AirCap with multicopters

Real Airship

To run the controller on a real airship, the vehicle needs an OpenPilot Revolution Flight controller or compatible and a companion computer capable of running neural networks and with wireless communication capability connected via USB. We are using an NVIDIA Jetson TX1 or TX2 with integrated wifi. ROS should ideally be run in a multi master setup using fkie_multimaster, with an instance of the MPC running on each vehicle.