POVNav: A Pareto-Optimal Mapless Visual Navigator

POVNav is a planning and control framework that uses a novel image-based local representation of the environment using existing image segmentation methods, to navigate a robot to the specified goal point or direction using only a monocular camera without relying on a map. Apart from the collision-free motion, it also shows selective navigation behavior (such as ``Do not walk on the grass!'') which is not possible with the occupancy grid representation. The navigation task is formulated as a visual servoing problem where the robot is able to directly generate efficient motion according to the visual features representing the navigability. The navigability is represented as a binary image generated from existing segmentation techniques, making it a handy plug-and-play suite for any segmentation method.

This paper has been accepted for publication in ISER 2023.

• Paper Link: https://arxiv.org/abs/2310.14065

Cite as:

@article{pushp2023povnav,
  title = {POVNav: A Pareto-Optimal Mapless Visual Navigator},
  author = {Durgakant Pushp and Zheng Chen and Chaomin Luo and Jason M. Gregory and Lantao Liu},
  journal = {arXiv},
  year = {2023},
  eprint = {2310.14065},
  archivePrefix = {arXiv},
  primaryClass = {cs.RO},
}

Requirements

• Any robot simulator that provides the control of it's linear and angular velocities (v, w), e.g. Jackal, Husky, Turtlebot e.t.c. Use the following instruction to setup Jackal simulator: sudo apt-get install ros-<distro>-jackal-simulator ros-<distro>-jackal-desktop ros-<distro>-jackal-navigation.
• Image Segmentation Method: Use any segmetation method of your choice, define the nevigabiliy vector and prodice binary segmentation image such that navigable segments gets a value '255' non-navigable gets '0'.

Required ROS topics

Subscribe	Description	Example
~odom_topic	name of topic that provides the robot's state.	"/odom"
~binary_segmented_image	name of the topic that provides the binary segmeted image based on the defined navigability.	"/ground_segmentation"

• It publishes linear and angular velocities to the "/cmd_vel" topic.

Setup

Install and run the dependencies

• Launch the jackal simulator (or any other UGV sumulator) with realsence camera on it. Verify that you are able to subscribe to the above mentioned topics.

export JACKAL_URDF_EXTRAS=/path_to_src/camera_urdf/realsense.urdf.xacro
roslaunch jackal_gazebo empty_world.launch 

• Setup segmentation module from depth2surface_normals_seg. Alternatively, you can run any image segmentation algoritms and implement Visual Horizon. Launch the segmentation node.

Setup POVNav repository

How to Build

Clone and build the repository to your workspace.

mkdir catkin_ws/src -p
cd catkin_ws/src

git clone https://github.com/Dpushp/POVNav.git

cd ..
catkin_make

Lanuch the sim_pov_nav node

roslaunch roslaunch pov_nav sim_pov_nav.launch 

Open rviz to visualise the topics and give goal to the robot

rosrun rviz rviz -d path_to_pov_nav_src/rviz/pov_nav_viz.rviz

Robot will navigate to the goal given from rviz.

Parameters Tuning

Run Dynamic Recongigure to change the parameters.

rosrun rqt_reconfigure rqt_reconfigure