Module: Camera Subscriber Node (CamSubNode)

CamSubNode is one of the most mature reusable nodes in ros_base. Its responsibilities are clear:

• Subscribe to RGB, depth, infra, and CameraInfo
• Convert ROS image messages into numpy.ndarray automatically
• Cache the latest results into readable attributes
• Provide lightweight visualization and recording when needed

1. Data exposed by the current implementation

After the node starts, commonly used attributes include:

• img: latest RGB image, numpy.ndarray
• img_timestamp
• img_shape
• depth
• depth_timestamp
• depth_shape
• rgb_info
• infra1 / infra2
• infra1_info
• infra_fx
• infra_baseline

That is why handlers and agents can write:

img = self.camera.img
depth = self.camera.depth
timestamp = self.camera.img_timestamp

2. Supported camera configurations

CamSubNode currently accepts either a string preset or a CameraProfile as config.

Built-in string presets include:

• realsense_d435i
• realsense_d435i_compressed
• realsense_d435i_align
• realsense_d435i_align_compressed
• realsense_d435i_infra
• zed_mini

One detail to note: although the constructor type hint mentions dict, the current implementation actually accepts only:

• str
• CameraProfile

3. Typical usage in attached mode

The entry script in SigLoMa-VLM registers it directly into nodes_dict:

nodes_dict = {
    "vlm_node": Robot2VLMBridge,
    "joystick": JoystickSDKNode,
    "camera": CamSubNode,
}

Then the manager uses a handshake rule to guarantee that the first frame has arrived:

self.add_handshake_rule(
    "Camera Stream",
    lambda: self.nodes["camera"].img is not None,
)

That way, by the time handlers.handle() actually starts, a valid image is usually already available.

4. Typical parameters

Parameter	Default	Description
config	"realsense_d435i_align"	Preset camera configuration
record	False	Whether to record video in a background thread
record_fps	20	Maximum recording frame rate
vis_rgb	False	Show an RGB window
vis_depth	False	Show a pseudo-color depth window
gamma	2.0	Gamma preprocessing
use_clahe	False	CLAHE local contrast enhancement

Recording behavior

If record=True, the current implementation will:

• Start a background thread
• Write images to an mp4
• Write timestamps into a text file

The output directory format is:

~/Data/rosbags/track_session_YYYYMMDD_HHMM/

5. Standalone debugging

The source file already provides a command-line entry:

python3 ros_base/nodes/camera/cam_sub_node.py \
    --camera realsense_d435i_align \
    --vis_rgb \
    --clahe

If you also want to inspect depth:

python3 ros_base/nodes/camera/cam_sub_node.py \
    --camera realsense_d435i_align \
    --vis_rgb \
    --vis_depth

6. Working with PoseProcessor

CamSubNode itself only acquires images and camera info. The actual conversion from:

• bbox + depth
• odom
• camera extrinsics

into world coordinates is handled by ros_base.nodes.camera.pose_processor.PoseProcessor.

In SigLoMa-VLM, this object is held by Robot2VLMBridge.pose_processor and is used to:

• Get the robot base position in the world frame
• Convert a selected target box into world coordinates

7. What it should and should not do

Good responsibilities:

• Subscribe to and cache the latest image data
• Lightweight preprocessing
• Temporary visualization
• Development-time recording

Poor responsibilities:

• Running heavy model inference inside callbacks
• Holding the whole vision business logic
• Mixing complex state machines into a single callback

Those parts should stay in agents and handlers.