Conversation
yuanknv
requested review from
MiguelCompany,
ahcorde,
hidmic,
karsten-nvidia,
mjcarroll,
nvcyc and
tfoote
ahcorde
left a comment
ahcorde
left a comment
There was a problem hiding this comment.
- Add CONTRIBUTION.md and LICENSE
- Add common linters
- Add full license header in file
| - [rcl_buffer_ws](https://github.com/yuanknv/ros2) workspace with `pixi` | ||
| - LibTorch (see [torch_buffer_backend prerequisites](https://github.com/yuanknv/torch_buffer_backend#prerequisites)) |
There was a problem hiding this comment.
can we place this information in ros2/ repositories ?
There was a problem hiding this comment.
This is now out of date since we already have the libtorch_vendor package inside the rosidl_buffer_backends
|
|
||
| ### Dependencies | ||
|
|
||
| - [rcl_buffer_ws](https://github.com/yuanknv/ros2) workspace with `pixi` |
There was a problem hiding this comment.
We should assume to use the ros2/ros2 but not the pixi-patched one.
We should also list out key packages like cuda_buffer_backend and torch_buffer_backend, in addition to the ros2/ros2 repo, which will be helpful as a tutorial to show what are being tested here.
| # 1. Build CUDA device backend | ||
| pixi run build cuda_buffer_backend | ||
|
|
||
| # 2. Build the demo | ||
| pixi run build robot_arm_demo |
There was a problem hiding this comment.
Let's not use pixi commands as they are not in the rolling branch in ros2/ros2.
Let's use plain colocon build commands in this tutorial.
There was a problem hiding this comment.
Changed the readme to use rosdep and colcon to build and verified in the pixi shell.
Thanks for the suggestion, all applied Co-authored-by: Alejandro Hernández Cordero <ahcorde@gmail.com> Signed-off-by: yuanknv <113960800+yuanknv@users.noreply.github.com>
3 participants
Description
Add an end-to-end robot arm demo to utilize the
rosidl_buffer_backends. The demo is a two-node ROS 2 application where a renderer node renders an SDF-based pencil-sketch robot arm animation entirely on the GPU via pure torch and publishes BGRA frames assensor_msgs/msg/Image, while a display node subscribes and presents them in an SDL2/OpenGL window.It demonstrates the
torch_buffer_backendwith CUDA zero-copy IPC against the default transport (no buffer backend). For the CUDA path, data stays on the GPU across processes. For the default path, frames are copied to host and go through standard ROS 2 middleware.The benchmark shows around 3-6x throughput improvement with zero-copy over the default path: ~117 vs ~36 fps at 1080p, ~91 vs ~21 at 1440p, and ~60 vs ~10 at 4K on an RTX 3090.
Fixes # (issue)
Is this user-facing behavior change?
Did you use Generative AI?
Yes. Claude (claude-4.6-opus) via Cursor was used to assist with creating an initial prototype version of the changes contained in this PR.
Additional Information
This PR is part of the broader ROS 2 native buffer feature introduced in this post.