[RFC] Getting feedback from base controller

[gerkey]

I have a situation in which my robot's base controller (using the terminology from this block diagram) is somewhat complex. It has internal logic by which it might determine that a given cmd_vel is impossible to execute (this robot has a steered caster setup with steering limits).

It's easy enough for me to have the base controller stop the robot in that situation, but it would be nice to propagate that failure information back into move_base somehow. It seems to me that accepting such failure feedback from the base controller is similar to how move_base already watches whether the global and local planners fail and then takes various actions (recovery behaviors, abort, etc.). For my particular application, I want to be able to stop pursuit of a goal if the base controller is ever asked to execute a velocity that it determines to be invalid.

Would others consider this a useful feature? There are probably a variety of situations in which the base controller might fail or refuse to move the robot, either due to software logic or hardware failures. Has anyone given thought to taking feedback from the base controller into move_base? I'm currently leaning toward adding an action interface to my base controller and then modifying move_base to optionally be able to send velocity commands there instead of doing a raw publish to cmd_vel. Other ideas on how to implement it?

[corot]

Hi Brian, you are clearly calling for move_base_flex (hope you attended the presentation at ROSCon! 😠).

Well, is not exactly what you need, but for sure can accommodate possible solutions way easier than current move_base. A simple option could be, assuming you are using BT or SMACH to control your robot through move_base_flex, to add a "watcher" task monitoring the progress of your robot. I case of "illegal" velocity commands, the robot will stop and the watcher will detect that the base is ignoring velocity commands and raise the appropriate error. Or he can directly listen for a diagnostics topic provided by the base, including some kind of "illegal command received" flag.

This sounds like a bit patchy... but we are doing exactly that to detect the non-very-infrequent case in which our robots get power cut due to safety violations.

Another (more elegant? 😳) option would be that the local planner plugins loaded by move_base_flex provide a second (optional) usage mode, in which they receive the plan and make all the controlling talking directly with the mobile base, instead of returning a velocity command at each iteration.

Happily, I'm right now redesigning the move_base plugins interface loaded on move_base_flex, so they can return a meaningful error code plus an optional message, instead of just true/false. You can take a look at the current status here. On move_base_flex_msgs actions we provide some common return codes plus ranges of values usable by the plugins developers. Interestingly enough, @DLu did something very similar (and also presented at ROSCon), so ideally we should find a common ground to merge our move_base_flex_core and his nav_core2.