Introduction

Visual navigation is the package responsible for stereo camera, Isaac ROS VSLAM, and point odometry. It contains multiple launch files for launching different configurations. The package integrates visual SLAM for odometry estimation and stereo vision for depth perception, bridging to PX4 for drone navigation.

Launch Files

The package includes several launch files for different configurations. The launch files don’t contain the full lists of parameters, for those consult the oness listed from the node section below. Generally though launch files are the preferred way of launching the nodes.

stereo.launch.py

Launches the stereo_publisher node with configurable parameters for the stereo camera setup.

vslam.launch.py

Launches the vSLAM pipeline without the stereo publisher, assuming camera input from elsewhere. This can be used for example when playing back a rosbag of a recorded flight data of depth camera. This does NOT start the vslam_message_transform node.

stereo_vslam.launch.py

Launches both stereo_publisher and vSLAM components together for integrated stereo vision and SLAM. This is the simplest option to use, as it launches both at the same time.

Nodes and their short explanations

Nodes included in visual_navigation:

stereo_publisher

Custom wrapper for the OAK-D pro camera. The default wrappers had all sorts of unnecessary features running which caused a brownout so we wrote a minimalistic wrapper (brownout could possibly be solved by using the Y adapter). Publishes stereo camera feed, and IMU data using the DepthAI (Luxonis) OAK-D camera. For the full list of launch parameters and other examples and API documentation refer to https://docs.luxonis.com/software/ros/depthai-ros/ (also includes how to use the other features of the camera)

vslam_message_transform

Transforms odometry messages from Isaac ROS Visual SLAM to PX4 VehicleOdometry format. It subscribes to the vSLAM odometry topic, performs coordinate frame transformations (from vSLAM axes to PX4 FRD), and publishes the transformed odometry for PX4’s EKF2. This node has no configurable parameters and uses default topics: subscribes to “/visual_slam/tracking/odometry” and publishes to “/fmu/in/vehicle_visual_odometry”. For more info on the VSLAM visit https://nvidia-isaac-ros.github.io/v/release-3.2/repositories_and_packages/isaac_ros_visual_slam/isaac_ros_visual_slam/index.html

point_odometry

Converts tracked object detections into vehicle odometry messages. This node processes Detection2DArray messages, extracts tracked objects, and computes odometry based on their positions. Subscribes to “detections” topic (vision_msgs/Detection2DArray) and publishes to “/fmu/in/vehicle_visual_odometry” topic (px4_msgs/VehicleOdometry).