Fleet adapter crash with no error when doing a multi floor planning

[koonpeng]

Bug report

Required information:

• Operating system and version:
  • ubuntu22.04
• OpenRMF installation type:
  • source
• OpenRMF version or commit hash
  • main
• ROS distribution and version:
  • humble
• ROS installation type:
  • binary
• Package or library, if applicable:
  • rmf_fleet_adapter

Description of the bug

When a patrol task is sent with waypoints in different floors, the fleet adapter exits silently with no error. The last message it prints is from

rmf_ros2/rmf_fleet_adapter/src/rmf_fleet_adapter/agv/FleetUpdateHandle.cpp

Lines 292 to 300 in 3c17991

	if (task_managers.empty())
	{
	RCLCPP_INFO(
	node->get_logger(),
	"Fleet [%s] does not have any robots to accept task [%s]. Use "
	"FleetUpdateHadndle::add_robot(~) to add robots to this fleet. ",
	name.c_str(), task_id.c_str());
	return;
	}

.

[fleet_adapter-16] [INFO] [1673946622.213851799] [tinyRobot_fleet_adapter]: Planning for [2] robot(s) and [1] request(s)
[ERROR] [fleet_adapter-16]: process has died [pid 1083490, exit code -11, cmd '/home/koonpeng/ws-osrf/cgh/install/lib/rmf_demos_fleet_adapter/fleet_adapter -c office/tinyRobot_config.yaml -n office/nav_graphs/0.yaml --ros-args --params-file /tmp/launch_params_3bx177v3 --params-file /tmp/launch_params_5ffcvn_3'].

Steps to reproduce the bug

office.tar.gz

Extract and start the custom scenario

ros2 launch office/office.launch.xml

then send a patrol request

ros2 run rmf_demos_tasks dispatch_patrol -p patrol_A1_L1 patrol_A1_L2

Expected behavior

Successfully plan a path, or fail gracefully if it cannot find one.

Actual behavior

Crashes with no error messages.

[Yadunund]

Upon investigating the building.yaml file, it seems

• Level fiducials were missing which are needed to derive the scale for the second level. Without this, waypoints on higher levels might have retained their pixel-coordinate values.
• several waypoint properties had their keys incorrectly modified (_L1 or _L2 were appended)
• Level 2F had same elevation as 1F

I've fixed these discrepancies and you can find the updated building yaml below.
office_fixed.zip

However it will be good to fix a bug that surfaced in GoToPlace where we're not checking if battery_soc > 0 before updating the finish state.
https://github.com/open-rmf/rmf_task/blob/7bc64b63ea163e5f90c70f76b372102f27440678/rmf_task_sequence/src/rmf_task_sequence/events/GoToPlace.cpp#L142-L145

[koonpeng]

Yeah, the map was hackily generated. Thanks for the fix.

Should I close the issue now or leave it open till the battery bug is fixed?

[mxgrey]

Let's leave it open until the task planning issue is fixed. There are no scenarios where it's acceptable for the task planner (and more broadly, the fleet adapter) to crash.