Custom RMF package development.

[rocker12121]

I want to make a custom rmf package that is very lightweight and only cares about traffic conflict detection and scheduling. Can i just use the rmf_traffic repository/package for that? What is the most appropriate way to do that?
Basically I want to use the core RMF functionality of traffic management in a distributed way.

Also, is there a way to run rmf_traffic_schedule node separately on each robot? if yes, how can I do that? Please guide.

[mxgrey]

What is the most appropriate way to do that?

What you're describing is the purpose behind rmf_traffic_ros2 which gets used internally by rmf_fleet_adapter.

In theory rmf_traffic_ros2 was meant to be used the way you're describing. In practice it's somewhat difficult to tie everything together because async programming in C++ is sort of inherently difficult. We wrap up the implementation of all the async planning and negotiation into rmf_fleet_adapter, exposing a much simpler API to integrate robots against.

I have some words of caution about building something directly against rmf_traffic_ros2 at this point in time:

1. The async programming will be difficult and error prone (see the complexity of rmf_fleet_adapter's implementation for reference).
2. In the next generation we're replacing the functionality of rmf_traffic_ros2 with mapf and bevy_impulse. We will be making a shim so that the rmf_fleet_adapter API can be used with these new libraries underneath, but we will not be making a shim for rmf_traffic_ros2 because its API doesn't align with the new capabilities that we'll be introducing in the next generation.
3. The next generation libraries will be much easier to use for pure traffic scheduling because they are designed to be far more modular.

Overall, what you want is something that you'll see better support for in our forthcoming work. You could use what's already available today, but it's likely to be deprecated by the time you finish getting it to work.

If you can elaborate on your use case and development timeline maybe I can give more guidance on which path makes more sense.

[rocker12121]

Got it.

One more thing.. Is there any way I can check the behavior of the robot when it loses connectivity during the mission?

[rocker12121]

What happens when a robot decomissions in the middle of a task and the idle behavior is true. And what is the best way to test it?

[mxgrey]

Is there any way I can check the behavior of the robot when it loses connectivity during the mission?

I may need you to elaborate on this question. I'm not sure which of the following you are asking:

1. "What is the behavior when the robot loses its network connection?"
2. "What API can I use to detect that the robot has lost its network connection?"

For (1) it will depend on how the fleet adapter for your particular robot platform is implemented, but in most cases the robot will continue carrying out the last command sent to it from RMF and then wait for its next instructions. From the RMF side, RMF will wait to issue the next command (s) until it has positive confirmation that the robot has reconnected and finished its last command. This may also mean that other robots who did not lose their connection will be delayed if they need to wait until the disconnected robot has finished its last command.

This situation happens very frequently with lifts since most robots will completely lose their network connection while they are inside a lift, so Open-RMF is designed to handle this gracefully.

For (2) there is no Open-RMF API for detecting whether a robot has lost its network connection since different robots have very different means for determining this.

[mxgrey]

What happens when a robot decomissions in the middle of a task and the idle behavior is true.

Decommissioning never affects an ongoing task. Ongoing tasks can only end by successfully finishing or by sending a cancel_task_request.

[mxgrey]

You can test the decommissioning behavior by running a demo and using the api_request script to send a commission request to a robot while it's performing a task.