Bug: refresh_camera_data return value ignored in camera_subprocess_func, causing stale frames to be written to SHM

[heuljupupup]

Bug Description

In instinct_onboard/ros_nodes/camera_base.py, the camera_subprocess_func
loop ignores the return value of camera.refresh_camera_data(). This can
cause stale frames to be re-written to shared memory with an incremented
frame_counter, making the main process believe a new frame arrived.

Location

instinct_onboard/ros_nodes/camera_base.py, lines ~227-242 (depth) and
~244-251 (color), inside camera_subprocess_func.

Root Cause

1. CameraBase.refresh_camera_data() returns bool indicating whether new
   data was acquired (True) or not (False).
2. RealsenseCamera.refresh_camera_data() only updates self._depth_data
   when frames.get_depth_frame() is not None. If the depth frame is missing
   from the frameset, _depth_data retains the previous frame.
3. The subprocess loop discards the return value:

   camera.refresh_camera_data()          # return value ignored
   if depth_buffer is not None:
       depth_data = camera.get_depth_image()
       if depth_data is not None:        # passes — stale data is not None
           depth_buffer[:] = depth_data
           depth_header.frame_counter += 1   # counter incorrectly incremented
           ...

Consequence

• When get_depth_frame() returns None, the previous frame is written to SHM
  again with a new frame_counter and timestamp.
• The main process (CameraProcessSpawner.refresh_camera_data) sees
  frame_counter changed and treats the stale data as a fresh frame.
• Downstream agents receive an out-of-date depth observation with an updated
  timestamp, which is indistinguishable from a real new frame.
• The same issue affects the color stream.

Trigger Condition

Rare: pipeline.wait_for_frames() succeeds but frames.get_depth_frame()
returns None (USB bandwidth jitter, IR projector transient, firmware sync
loss).

Suggested Fix

Option A — respect the return value in the subprocess loop:

got_new = camera.refresh_camera_data()
if depth_buffer is not None and got_new:
    depth_data = camera.get_depth_image()
    ...

(Note: got_new is True if either depth or color was acquired, so this
is a coarse fix. A per-stream status would be more precise.)

Option B — invalidate the cached data when no new frame arrives:

# In RealsenseCamera.refresh_camera_data, when depth_frame is None:
self._depth_data = None

This makes the existing if depth_data is not None check correct, but
changes the "keep last frame" behavior that single-process consumers may
rely on.

Environment

• Ubuntu 22.04, ROS2 Humble
• Jetson Orin NX (ARM, weak memory order — the memory barriers in this file
  suggest ARM is a deployment target)