Consideration of using LCM?

[shuoyangrobotics]

Consideration of using LCM?

[ShuoYangRobotics]

Hi Shengzhi,

Thanks for your question. On hardware, ROS is not involved in critical communication. ROS only publish debug messages. The robot and the controller directly talk via unitree SDK's UDP class. This is the most direct communication channel. You cannot do any better. The LCM is a wrapping layer over the UDP class so theoretically it introduces a small delay. Using UDP communication the control frequency can reach around 400Hz-500Hz, which is enough to achieve torque control with reasonable performance. The controller is not running real time but I think if your robot does not run at 5m/s, this won't affect performance.

In Gazebo simulation, yes, gazebo and the controller talk via ROS. But we can slow down the simulation speed to make sure everything is still synchronized.

[edward9503]

Hi Shuo,

thanks for your reply. I've attached the A1 software guide, and I would like to quote the section 4.4.1 in page 15.
A1 Software Guide v2.0-en.pdf

In my understanding, one ROS node acts like one process in pc. It is commonly known that ROS is not real-time guaranteed. Although you are using ROS not for the critical communication, your controller runs as a non-real-time Ros node and the UDP instructions are set directly inside this node, which thus cannot ensure the real-time performance on controlling the robot. As mentioned in the software guide, by using LCM, the non-real-time ROS nodes won't affect the real-time process of controlling robot. Therefore, it seems that using LCM could help to achieve the real-time.

How do you think?

Best

[edward9503]

Hi @ShuoYangRobotics ,

any feedback on this? I've also asked the technical support from Unitree. They also recommend using the LCM instead of direct UDP. So maybe this could help to improve the real-time performance. How do you think?