uncertain_turtlebot_navigation

A turtlebot navigation simulation in Gazebo using uncertainty maps.

The uncertainty maps used in this demo environment are constructed according to:

F. Verdoja, J. Lundell, and V. Kyrki, “Deep Network Uncertainty Maps for Indoor Navigation,” arXiv:1809.04891 [cs, eess], Sep. 2018.

Please refer to that paper for more details about the map building process.

Use

• First, you need to start a simulated turtlebot in an empty gazebo environment:

  roslaunch uncertain_turtlebot_navigation turtlebot-sim-empty.launch

  If you don't want the gazebo gui to launch, add gui:=false.

• Then, launch the navigation you prefer:

  • For normal slam navigation:

    roslaunch uncertain_turtlebot_navigation slam_navigation.launch

  • For uncertain navigation:

    roslaunch uncertain_turtlebot_navigation uncertain_navigation.launch

    The default uncertainty map in this case will be the one using a Laplace model, in alternative you can use the argument uncertainty:={laplace|gaussian|mcdropout} to select the uncertainty map to be used.

• Now you can either plan by setting navigation goals through rviz or by executing a set of sample navigation goals by using:

  roslaunch uncertain_turtlebot_navigation execute_navigation_goals.launch

• Once you have run the navigation and stored the trajectories in the folder /data/trajectories, you can analyze the results, that is compute the number of collisions that occured, by running the script calculate_collisions in the matlab folder.