Custom request in office.world

Yadunund / rmf_test_scenarios

Launch files to fleet adapters

0 stars 0 forks source link

Custom request in office.world #3

Open Kevinskwk opened 4 years ago

Kevinskwk commented 4 years ago

The magni fleet adapter crashes during testing.

Requests sent one after another from rviz panel:

Loop request:
- Start waypoint: coe
- End waypoint: hardware_2
- Repeat count: 10
Loop request:
- Start waypoint: hardware_1
- End waypoint: coe
- Repeat count: 10

Workspace:

rmf_core: develop
rmf_demos: develop
rmf_schedule_visualizer: master
traffic_editor: master
rmf_assets: master

Attached terminal output office_test.txt

Yadunund commented 4 years ago

@Kevinskwk do you have a screenshot of Rviz?

Kevinskwk commented 4 years ago

Sorry I forgot to take that... But the situation was both robots just went out of the doors (core door and hardware workshop door, one robot from each). And in rviz, it seems that a viable path without conflict has already been found. And it was also the same time that the fleet adapter crashed.

mxgrey commented 4 years ago

The error indicates that the planner believes there is no viable path between the waypoints you asked for, regardless of any other vehicles. Obviously the planner is mistaken, but I can't determine for certain the reason why with only the information that got printed out. I do have a theory for what may have caused it, and I've added some more error printout in the latest changes in case this happens again. Hopefully the next time this happens my hypothesis can be confirmed, and then maybe I can do something about it.

mxgrey commented 4 years ago

For the record, I've tried to recreate the crash in this scenario, and it's just not happening for me, so I'll assume that it happens intermittently. If anyone is able to reproduce it consistently, then please try to recreate it with the latest rmf_core:develop changes.

mxgrey commented 4 years ago

Oh never mind, I ran it again and this time it did produce the crash, and the crash confirmed my hypothesis. I'll start working on a fix for this tomorrow; thanks for the bug report!

mxgrey commented 4 years ago

The latest changes should suppress this issue, although the root cause is not currently fixed. If you run this experiment again, let me know if you see the same error.

I will note that because of the negotiation memory leak, there's a good chance that this scenario will max out your RAM before very long.