diff_opt_tracking

This repository is associated with our RAL submission paper: "Differentiable-Optimization Based Neural Policy for Occlusion-Aware Target Tracking".

Dependencies

• JAX
• bebop_simulator
• gazebo2rviz (If you need the RViz visualization)

Installation procedure

After installing the dependencies, you can build our propsed MPC package as follows:

cd your_catkin_ws/src
git clone https://github.com/Houman-HM/diff_opt_tracking
cd .. && catkin build
source your_catkin_ws/devel/setup.bash

Running the algorithm

In order to run the MPC for the target tracking, follow the procedure below:

In the first terminal:

roslaunch diff_opt_tracking random_world.launch

This launches a random environment with various obstacles in Gazebo.

In the second terminal:

rosrun diff_opt_tracking planner.py

You can now publish velcoities on "target/cmd_vel" topic to move the target. As the target moves, the robot would follow it while avoiding occlusions and collisions.