Viability of using RCLCPP as the client library for low resource arm based linux systems #398
Comments
|
Hi @coalman321. Give it a try. The interface for rclcpp is rcl, so there should be no other dependencies (potentially, the type conversion on xrce-dds-client support might be an issue). Mainly you have to watch out for total memory consumption as well as memory fragmentation with RCLCPP. You could make some measurements on a normal linux desktop with ROS 2 Galactic and rclcpp - with your C++ application and check with tools like valgrind, how much (static/dynamic) memory is actually used. This might give you an indication, if 256MB RAM is sufficient. My question would be - with such a powerfull machine with large memory, why are you not using ROS 2 directly, as you have Linux with Preempt-RT patch already running? Functionally RCLCPP and RCLC are compliant (RCLC is feature complete: creation of nodes/publisher/subscription/service/client/parameters/...)The RCLC has a (I would argue) a more advanced executor, it offers more deterministic features for real-time behavior. But if you don't need that in your application, start with default RCLCPP Executor. This is the only major difference between RCLCPP and RCLC. Just noting, the rclc-executor pre-allocates memory in the configuration-phase. At runtime there is no dynamic memory allocation. This is not the case for the RCLCPP-Executor (which is using STL-containers). |
|
Our biggest issue with RCLC right now is that it doesnt seem to have any real way to deal with C++ and its classing system, something our software is based around. We attempted to build minimal versions of ros2 for the platform, but were unable to get any real traction. Many system dependencies that ROS expects to have access to would not cooperate nor build properly, largely due to the older compiler and kernel being used by the system (something we do not have control over). That aside, I did make an attempt, by adding RCLCPP into the mcu_ws as well as any other dependencies it has listed and the build does start as expected, but seems to be failing due to missing C++ type support. have not yet had a chance to work out out why, other than the build fails on libstatistics_collector due to missing some of the C++ builtin_interfaces. This seems strange as C++ typesupport is present, and seems to build okay, but isnt producing results. These are the current build args for the generate_lib build.sh script: And this is the list of the current packages in mcu_ws: |
|
Okay, so that was a failure on my part. I had forgotten Colcon copies all executables into the install directory the uses the copy when the executable is run. I updated my micro_ros_setup build and it no longer fails at libstatistics_collector. Now its failing at RCLCPP. the current error is as follows: It looks like the context manager for micro_ros's RCL may not be strictly compatible with the context manager RCLCPP is expecting. The direct cause of this though seems to be a flag not being set for the compilation of RCL. |
|
Tracking the resolution of this compilation issue with the RCLCPP group. ros2/rclcpp#1803 |
|
I have also documented first steps for a C++ wrapper around RCLC here: PR ros2/rclc#199 and issue ros2/rclc#126 |
|
Hi @coalman321 cc @pablogs9 Yes, you need to define
This is an issue with the micro-ros/rcl package not the ros2/rcl package: |
|
ROS 2 Galactic requires C++17 compiler: Does your compiler support that? |
The compiler is an armv7 build of GCC, version 7.3.0, which does support the C++ 17 standard. I am not 100% sure on the python 3.6 support, though I believe the answer is yes. The armv7 version is not currently installed on the build machine though.
Colcon does show the message during compile of RCL. |
Per this comment, I have some new findings. This weekend, I was able to cross compile up to rclcpp using a stripped down ros installation for the platform. I am still curious about the application for micro_ros, and am willing to re-try the compilation process with both a valid sysroot, and toolchain setup. Although, in this situation, I doubt that it will solve the missing compile flags being an issue, as it was not an issue for normal micro_ros builds. ` |
Hi, I have a question about the viability of using RCLCPP in place of RCLC for low power / resource constrained arm systems. So far I have a working version of micro_ros for the platform, as well as a working colcon.meta and toolchain, but was curious if it would be possible to use RCLCPP as the client library for the micro_ros client. Given our use case is already C++ on the device, using RCLCPP seems to be a better fit. Especially rather than than having to write a significant wrapper around RCLC. Would this switch even be possible, or are the two client libraries functionally different enough to a point where it would not work? Thank you for your help! (Sorry if this is the wrong place to pose this question)
The text was updated successfully, but these errors were encountered: