Using motoman in fake simulation?

[riv-tnoble]

Hi,

I'm using motoman to stream points to a robot and want to do the same in a fake simulation so I can test code quicker in an environment as close to the real thing as possible. I've tried using localhost as the robot IP, hoping that it'd play nicely with the fake controllers we have set up via MoveIt, however, I get the same error from every motoman node:

[ERROR] [1655814964.985871203] [/fake/motion_streaming_interface]: Failed to connect to server, rc: -1. Error: 'Connection refused' (errno: 111)

I believe the hardware controller is expected run a MotoROS server to provide the interface to control the robot to motoman, so I'm wondering if the connection is failing since there's no server to communicate with in fake? Fake controllers (e.g. MoveIt's) do exist though, so I'm wondering if there's a way to use them with motoman.

All of the topics I've seen online discuss using motoman with the real hardware controller, so I'm a little unsure where to start (or if it's even the best idea). I'd much appreciate any help and I'm open to suggestions 😄

[gavanderhoorn]

I've tried using localhost as the robot IP, hoping that it'd play nicely with the fake controllers we have set up via MoveIt

I'm not sure how that was supposed to work. MoveIt knows nothing about motoman_driver or the other way around.

motoman_driver interfaces with a MotoROS server application -- typically running on a Yaskawa Motoman robot controller.

MoveIt's "fake controllers" do not run a MotoROS server application, nor do they expose the same functionality. In fact, they're not even really controllers, as it's just a nice way to perform some kinematic playback of trajectories (ie: trajectory -> state -> visualisation).

So no, they cannot be used with motoman_driver.

This is not unique to motoman_driver btw: almost all ROS robot drivers can only be used with the actual physical hw, or a faithful simulation, such as often provided by robot OEMs in their programming / simulation environments.

Unfortunately, MotoSim does not support running MotoPlus applications at this time, so that wouldn't work either.

There is a non-MotoPlus-based version of the MotoROS server application, created by a community member (@tingelst): ra-mtp-ntnu/moto. Technically it's a Python implementation of Simple Message, with a couple of scripts to implement some basic server functionality which mimics MotoROS to a certain degree.

I've used that locally for testing motoman_driver when I don't have my real hw around.

But, and this is a fairly large but: this only gets you the basic, minimal Simple Message server application. It does not simulate anything really, nor does it behave exactly the same as MotoROS.

And I believe in your case, you've changed how motoman_driver behaves (ie: point streaming), which takes you even further away from the standard behaviour. You'll have to check whether whatever ra-mtp-ntnu/moto offers is compatible.

---

All of this assumes of course you need to 'simulate' at this level.

If all you're after is seeing a robot move in a UI, skip motoman_driver, use MoveIt's 'fake controllers' and call it a day.

But this:

so I can test code quicker in an environment as close to the real thing as possible

makes me believe that's not the case.

[riv-tnoble]

Thanks for the quick response @gavanderhoorn.

You're right, the benefit for us would be more than just visualisation (though that would still be useful as we don't have it currently). The repo you linked looks interesting. I'll have a look over the code and through our own modifications to the driver. Given what you've said though, it sounds like the best option would be to use MoveIt for the visuals, and write some automated tests that we can run easily on the hardware for speed of development 👍

[gavanderhoorn]

it sounds like the best option would be to use MoveIt for the visuals, and write some automated tests that we can run easily on the hardware for speed of development +1

well it would make a very nice contribution if you could create a MotoROS-like ra-mtp-ntnu/moto environment.

That would help a lot with debugging and system simulation.

[riv-tnoble]

That does sound nice. A couple of questions (as I'm a little unfamiliar with the lower level controller details):

1. You mentioned that the moto repo is incomplete. Is there some kind of complete API spec I could look at? I've found this document, which looks useful, although some of the details are a little vague. I've noticed the code in the MotoPlus directory of the repo too. Is that the source for the server running on the hardware?

2. RViz visualisation would be good, but I'm not 100% sure what that would involve. I'm hoping Rviz just updates the robot model visual based on e.g. the joint state topic, so that a mock State Server Task just needs to provide that feedback. Is that correct?

[gavanderhoorn]

1. I've noticed the code in the MotoPlus directory of the repo too. Is that the source for the server running on the hardware?

that is indeed the source of MotoROS.

You mentioned that the moto repo is incomplete. Is there some kind of complete API spec I could look at?

The document you found does give a high-level description of the behaviour of MotoROS. I wouldn't necessarily call it an API spec though.

Unfortunately I fear any design documentation besides the .pdf "has been lost to time", although we do have one of the main developers still around as a maintainer here (@EricMarcil).

I'd suggest we continue this discussion over at ra-mtp-ntnu/moto#15, where @tingelst just responded he'd like to expand/finish the server implementation he has there as well. We should be able to come up with the needed feature set and behaviour between the three of us.

RViz visualisation would be good, but I'm not 100% sure what that would involve. I'm hoping Rviz just updates the robot model visual based on e.g. the joint state topic, so that a mock State Server Task just needs to provide that feedback. Is that correct?

I would keep all of this completely separate.

Either you use MoveIt's fake controllers and everything 'will just work' (from a visualisation perspective) without any part from motoman_driver, or you create a 'fake MotoROS' implementation, still use the driver and then MoveIt and RViz would just 'see' a real robot (which is in fact a Python based Simple Message server).