X_LINK_ERROR

Copetags67

I have tried to get a custom ros2 launch file running to start up a pipeline of (Stereo depth, IMU data, RGB, and my segmentation masks from my own segmentation model) on your Luxonis OAK-D-Lite FF. I am running the ros-jazzy-depthai-ros-v3 on a raspberry pi 5 with Ubuntu 24.04. I can get the depthai_ros_driverv3 driver.launch.py to work every time, and have then made my own launch script, where i also try to launch my segmentation model converted to the .rvc2.tar.xz format. It will run the first time i startup my system from scratch, but all following times it will not work and i get this XLINK_ERROR that also tells me to "please report to developers" i have uploaded the crash file in a .zip format

home-robot-cache-depthai-crashdumps-23ce4fccbed804cfa3df53e0810c3813963c5fa6-crash-dump-2026-07-25-09-46-41.zip

16kB

Any idea what the problem might be? I have been at this for a few days now

Error in terminal:
robot@robot-desktop:/ros2_ws$ ros2 launch robot_bringup robot.launch.py

[INFO] [launch]: All log files can be found below /home/robot/.ros/log/2026-07-25-09-46-18-829940-robot-desktop-997

[INFO] [launch]: Default logging verbosity is set to INFO

[INFO] [tile_segmentation_node-1]: process started with pid [1000]

[INFO] [mask_to_costmap_node-2]: process started with pid [1001]

[mask_to_costmap_node-2] [INFO] [1784972780.072866224] [mask_to_costmap_node]: Lookup table built: 340305 / 409600 pixels map to ground plane (height=0.3m, pitch=25.0deg)

[mask_to_costmap_node-2] [INFO] [1784972780.177460038] [mask_to_costmap_node]: mask_to_costmap_node ready.

[tile_segmentation_node-1] [INFO] [1784972780.315847319] [tile_segmentation_node]: Building unified DepthAI pipeline (RGB + Depth + IMU + NN)...

[tile_segmentation_node-1] [2026-07-25 09:46:25] [depthai-nodes] [WARNING] [depthai-nodes.depthai_nodes.node.parser_generator] Segmentation based model detected with RVC2 device. Parsing will be done on the host machine.

[tile_segmentation_node-1] [INFO] [1784972785.897193366] [tile_segmentation_node]: Resolved depth algorithm: Algorithm.STEREO

[tile_segmentation_node-1] [INFO] [1784972785.898464873] [tile_segmentation_node]: Resolved depth config: PresetMode.FAST_ACCURACY

[tile_segmentation_node-1] [INFO] [1784972786.890511006] [tile_segmentation_node]: Pipeline started. Publishing /vision/rgb, /vision/drivable_mask, /vision/depth (aligned to RGB), /vision/imu/data

[tile_segmentation_node-1] [2026-07-25 09:46:28.574] [depthai] [error] Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_4__out' (X_LINK_ERROR)'

[tile_segmentation_node-1] [2026-07-25 09:46:28.575] [depthai] [error] Communication exception - possible device error/misconfiguration. Original message 'Couldn't read data from stream: '__x_7_depth' (X_LINK_ERROR)'

[tile_segmentation_node-1] [1944301011C1BC5A00] [4.1] [1784972790.645] [host] [warning] Closed connection

[tile_segmentation_node-1]

[tile_segmentation_node-1] [1944301011C1BC5A00] [4.1] [1784972790.645] [host] [warning] Attempting to reconnect. Timeout is 10000ms

[tile_segmentation_node-1]

[tile_segmentation_node-1] [2026-07-25 09:46:33.775] [depthai] [error] Device with id 1944301011C1BC5A00 has crashed. Crash dump logs are stored in: /home/robot/.cache/depthai/crashdumps/55087b96dfa2e2613cee460a3a1c3e468145ff7f/crash_dump_2026-07-25_09_46_33.tar.gz - please report to developers.

Oskar-Luxonis

Hey Copetags67 - thanks for the crash dump. The X_LINK_ERRORs are a consequence of the OAK device crashing during pipeline startup, rather than the root issue.

One thing to check first is the SHAVE allocation for your segmentation model. In this combined RVC2 pipeline, the model needs to fit alongside StereoDepth and the other nodes. Please make sure the .rvc2.tar.xz model is built for no more SHAVEs than the pipeline has available (in your case, likely 7 or fewer), or use an NNArchive/superblob variant configured for the available SHAVEs. More details are here: https://docs.luxonis.com/software-v3/ai-inference/inference/#setting-model-shaves

After updating the model, could you try launching the unchanged pipeline a few times? If it still crashes, please share your tile_segmentation_node pipeline code and the model archive/configuration so we can reproduce the exact camera + StereoDepth + NN setup.

Thanks,
Oskar

Copetags67

Oskar-Luxonis

I have set the NN to run on 6 shaves, and the problem is still present. It works the first time and then fail all following times until total restart and rebuild of my raspberry pi and docker container, respectively. My setup also includes streaming IMU readings so it is StereoDepth, IMU, RGB image, and NN. Earlier, I have gotten an error of the camera not having enough shaves to supply all nodes, but that does not seem to be a problem anymore and have not been for some time now.

Here is the tile_segmentation_node.py, path to Archive removed intentionally:

tile-segmentation-node.bin

#!/usr/bin/env python3
import rclpy
from rclpy.node import Node
from sensor_msgs.msg import Image, Imu
from cv_bridge import CvBridge

Error previewing file. It may have been deleted, or the provided file ID is invalid.

Is it possible to share the model somewhere else that is not this publicly available online forum to keep my model private?

Oskar-Luxonis

Copetags67 - yes sure, please write an email to oskar.sonc@luxonis.com and we can continue there.

Best,
Oskar