Running Oak-D-Lite RGBD Camera (with Mobilenet Detections) Causes Create3 100% CPU for 25 minutes #541
-
How are you connecting to your Create 3?Wi-Fi (ROS 2) Computer(s) Model(s) and Operating System(s)Ubuntu 22.04 in Docker over Raspberry Pi 5 PiOS Bookworm Which version of ROS 2 is installed on your computer?Humble Which firmware version is installed on your robot?H.2.4 Which RMW is your robot running?FastRTPS Does your robot have an assigned namespace? If so, please share.No response Is the robot connected to a network? If so, what is the network type?Yes: 2.4GHz Home WiFi Are there multiple Create 3 robots connected to your network?No Is multicast enabled?Yes What is the Adapter Board's USB/BLE Toggle currently switched to?Bluetooth (default) Describe your question.I have seen occasional 100%CPU when starting up new ROS 2 nodes before, but it seems running the Oak-D-Lite RGBD camera publishing RGB video, Stereo Depth "video", and MobileNet detections (with or without RTABmap, with or without rviz2 subscribing) cripples my Create3. This maxing out the CPU(s) was mentioned in the ROScon, and appears in several issue reports - is my plan for using the Create3 with the Oak-D-Lite for localization/navigation doomed? |
Beta Was this translation helpful? Give feedback.
Replies: 4 comments 3 replies
-
Test running camera for whole playtimeInterestingly, I can launch the camera (no RTABmap for this test) and still drive Create3-Wali slowly without crashing the Create3. Here is
The Create3 is reporting max 100% CPU but only 94% average load which appears to support responding to /cmd_vel topics acceptably: Note detecting chair and person classes with MobileNet
Disparity/Depth is being published but I couldn't get the image_view disparity_view or stereo_view to display the msgs. Rviz2 displays it the depth view, but it was not running in this test. My launch command was: and the camera.yaml file: The Raspberry Pi5 load averaged around 2 for this test (2 equals 50% of Pi5 CPU, 4=100% of all four cores): Which is drawing 8.7W total for the Oak-D-Lite and Raspberry Pi5:
Wali_node.py decide Wali had had enough playtime and commanded him back to his dock: He logged the session at 1.7 hours long (instead of his normal 2.8 hours w/out the camera and image processing: This is the camera command line output: Also contributing to success: |
Beta Was this translation helpful? Give feedback.
-
|
Besides the CPU usage, what surprised me is that your first post shows a very high RAM usage: Does your create 3 and the laptop need to communicate? Because you could restrict the create 3 to communicate only with the Raspberry Pi. Alternatively, you can try to use cyclonedds rather than fast-dds. Remember that if you change middleware, you need to change it everywhere (robot, raspberry pi, laptop) |
Beta Was this translation helpful? Give feedback.
-
|
@alsora Thank you for reviewing my post.
I'm not sure I understand "Does your create 3 and the laptop need to communicate?"
Do you mean the "Discovery Server" configuration described on the TurtleBot4 manual?
I am hesitant (due to my severe lack of knowledge) to "go back to cycloneDDS", in light of statements when Humble was released: From discourse.ros.org:
This does seem relatively easy to try, although I am not sure I have a specific reference test case measurements to know which is better. |
Beta Was this translation helpful? Give feedback.
-
Ok, from your use-case it looks like you indeed need to communicate between all the 3 entities (robot, raspberry pi, laptop).
That statement comes from the CEO of the company that develops Fast-DDS, so it may be a little bit biased. |
Beta Was this translation helpful? Give feedback.
-
|
Very interesting - (Business is a reality that drives everything - I'm along for the fun...) I did a read through of the turtlebot4 discovery server setup, and the user pc setup. Their setup allows access to the Create3 topics on the user pc. It sounds like it might be a bit of a pain when starting up a new node on my Pi5 controller - says I have to restart some server or servers, but that would allow rviz2 to see the Create3 generated /odom and /tf topics. The only thing it does not explain is how to access the Create3 logs in that configuration. Additionally, I would need to understand the turtlebot-xxx scripts and directory names to create3-xxx and directory names as I really do not want any Turtlebot dependancies, and while their code is "Open Source", it just feels weird to me to see turtlebot in my path, scripts, or node names. I read some of the eprosima fastdds documentation, but would have to rely on the turtlebot stuff to get it flying. I'm going to change WalI to the cycloneDDS "Simple Discovery" configuration and see if I get enough improvement. |
Beta Was this translation helpful? Give feedback.
-
|
@slowrunner to answer the question about the discovery server, here is a link to instructions on that: https://iroboteducation.github.io/create3_docs/setup/discovery-server/ The Turtlebot 4 configuration for a discovery server is a bit different. |
Beta Was this translation helpful? Give feedback.
-
|
New FW to test |
Beta Was this translation helpful? Give feedback.
Test running camera for whole playtime
Interestingly, I can launch the camera (no RTABmap for this test) and still drive Create3-Wali slowly without crashing the Create3.
Here is
The Create3 is reporting max 100% CPU but only 94% average load which appears to support responding to /cmd_vel topics acceptably: