Isaac ROS ESS disparity map incorrect

I’m following the Isaac ROS FoundationPose tutorial as described in this guide.

Steps 1-3 worked well:
Step 1: Isaac ROS ESS tutorial - Completed
Step 2: Isaac ROS RT-DETR tutorial - Completed
Step 3: Isaac ROS FoundationPose Quickstart Guide - Completed

However, when I launched the final script:

ros2 launch isaac_ros_foundationpose isaac_ros_foundationpose_isaac_sim.launch.py

I found that the disparity and depth images were not published correctly.

The /disparity topic contains repetitive values like:

header:
  stamp:
    sec: 28
    nanosec: 716668164
  frame_id: sim_camera
image:
  height: 576
  width: 960
  encoding: 32FC1
  data: 
    - 0
    - 0
    - 128
    - 191
f: 478.90576171875
min_disparity: 0.0
max_disparity: 2147483648.0
delta_d: 0.0

Disparity values appear invalid (repeating 0s and unrealistic numbers).
max_disparity is set to 2147483648.0, which seems incorrect.
delta_d is 0.0, meaning no disparity steps were calculated.
The ESS node is running, but the disparity output seems wrong.

Possible Cause?
My scene in Isaac Sim has a white ground plane.
I suspect that the lack of texture on these surfaces is causing stereo disparity matching to fail.

The scene is as below,

And this is when I ran the ESS node,

I ran the above ESS node with threshold = 0.35

  • Other nodes (RT-DETR, etc.) seem to be running correctly.
  • The only issue I could find is in the /disparity and /depth topics, where the values look incorrect.
  • I suspect that incorrect depth estimation is causing FoundationPose to fail because FoundationPose depends on accurate depth input.
  • Could the white surfaces (low-texture issue) be the reason for disparity failure?
  • What is the best way to improve disparity estimation in this setup?

+, as I published 4 topics in Isaac sim

/front_stereo_camera/left/camera_info,
/front_stereo_camera/left/image_rect_color,
/front_stereo_camera/right/camera_info,
/front_stereo_camera/right/imgae_rect_color.
  • Can I publish depth data from Isaac Sim and feed it directly into FoundationPose after resizing?
  • Would this be the best solution to avoid stereo matching issues?