Transformer-based Motion Model for Robust Target Tracking under Intermittent and Noisy Measurements

Requirements

• Run pip3 install -r requirements.txt to install all the required packages. You also need the following repositories.
• Reef estimator simulation.
  • You need to make sure all the submodules are up to date and tracking master branch. Do git submodule update --init --recursive to update the submodules.
  • To track master for all submodules, type git submodule foreach --recursive git checkout master
  • Note: to automatically change the number of turtlebots, inside the sim_helper repo, type git remote set-url origin https://github.com/andrespulido8/sim_helper.git. Then git checkout mml`
  • If you want to run only hardware then you only need reef_estimator
• Turtlebot packages for the turtlebot simulation. For both the turtlebot3 and the turtlebot3_msgs, checkout the noetic branch.
• Andres turtlebot PID controller for turtlebot controller.
• RosFlight (included in the Reef estimator simulator) for quadcopter autopilot.

Sim Usage

1. To adjust the number of simulated vehicles modify the ./launch/launch_sim.launch file inside the sim_helper repository.
2. Change the spawn_turtles argument inside the previously mentioned launch file to robot0.
3. In the sim_helper repository run python scripts/Master.py to start the simulation.
4. Wait a few seconds until Autopilot ARMED and RC override active are printed and then in another terminal run roslaunch mml_guidance mml_sim_estimator.launch from the launch directory. NOTE: To visualize the particle filter and the motion model, run roslaunch mml_guidance visualization.launch instead of mml_sim_estimator.launch.

Hardware Usage

To run only the turtlebot, do roslaunch mml_guidance turtlebot_hardware.launch. To run only the quadcopter, do roslaunch mml_guidance track_hardware.launch. To bag data during hardware experiments run roslaunch mml_guidance bag_hardware.launch prefix_name:="<insert prefix>"

Motion Model Learning (Neural Network)

The files needed to run the NN, the training datasets, and other supplemental material are located in this DropBox. You should move the csv to mml_guidance/scripts/mml_network/ and the .pth weights to mml_guidance/scripts/mml_network/models/. The DropBox also has the data used in the results of the letter.

Train

To turn off the Gazebo GUI to make the sim faster, change the argument gui to false in camera_multirotor.launch inside the sim_helper package from REEF github

Contributing Guide

To make changes to this repo, it is recommended to use the tool pre-commit. To install it, run pip3 install -r requirements.txt inside this repo, and then install the hooks specified in the config file by doing pre-commit install. Now to run it against all the files to check if it worked, run pre-commit run --all-files.

Profiling

Run roslaunch mml_guidance mml_sim_estimator.launch and then pprofile --format callgrind --out guidance.pprofile /home/andrespulido/catkin_ws/src/mml_guidance/scripts/guidance.py __name:=drone_guidance. This will run the profiler and save the results in the directory where the command was called.

Extra

Install tmuxinator to easily run the sim. Use sudo apt-get install -y tmuxinator. Then to run the tmuxinator command in the /mml_guidance directory. You can change the layout of the sessions with ctrl-b space