hunavis

Utilities for HUNAVsim, NAV2, and VISualization

Nodes

• people_visualizer: Visualizes detected humans as markers in rviz
• robot_pose_publisher: Publishes robot pose
• tf_keyboard_publisher: Publishes a transform that's adjustable using keyboard

Launch files

• hudet.launch.py: Starts human detection using a Zed2 camera.
• map_server.launch.py: Starts the map server without other Nav2 functionalities
• mars.launch.py: Starts a HuNavSim simulation in Gazebo
• tb3_custom_sim.launch.py: Starts Nav2 with basic robot navigation functionality

Suggested real-world setup for static Zed2 camera(s)

(TODO: Setups involving multiple cameras, which would mostly require repeating the steps below, each time specifying a different camera name)

1. Set up Zed2 camera(s) in the environment, with camera launch arguments defined in <zed_launch_args_file> (example) along with zed node and tf publisher parameters defined in <zed_and_tf_params_file> (example).

2. Create a map of the environment and save it to <map_path>

3. Launch map server, and optionally note the approximate positions of Zed2 cameras

   ros2 launch hunavis map_server.launch.py use_simulator:=False map_path:=<map_path>

   • empty_room.yaml is an example of <map_path>
   • Tip: Use the 2D Pose Estimate feature in rviz to set the camera pose on the map

4. Launch human detection

   ros2 launch hunavis hudet.launch.py use_simulator:=False zed_launch_args_file:=<zed_launch_args_file>

   • zed_launch_args.yaml is an example of <zed_launch_args_file>
   • If this is the first time deep learning models are run on the camera, the Zed SDK will begin to optimize them. Optionally, follow instructions here to optimize the models manually. For example, the following optimizes all the models that come with the camera:

     ZED_Diagnostic -aio

5. Run tf publisher node to adjust camera pose with respect to the map.

   ros2 run hunavis tf_keyboard_publisher --ros-args --params-file <zed_and_tf_params_file>

   • Optionally, <zed_and_tf_params_file> can be updated with the fine-tuned tf