DroneKit has no method for detecting command failure - do we need one?

[hamishwillee]

Currently our instructions suggest that you change the mode like this:

v.mode = VehicleMode("AUTO")
v.flush() # After this, write to vehicle has been made

The problem is that after this code the write has succeeded, but the mode may not actually have been changed (and if you read back v.mode.name it may well be the old name). A better coding paradigm is to not proceed to the next step until all your conditions are met:

print " Get ready to take off" 
v.mode = VehicleMode("AUTO")
v.armed= True
v.flush() # After this, write to vehicle has been made, but no guarantee change succeeded.
while not v.mode.name=='AUTO' and not v.armed==False and not api.exit:
    print " Waiting for AUTO and ARMED..."
    time.sleep(1)
print "Ready to take off"

Note: I will change our docs and examples to show the above.

There are two problems with this. Firstly it involves polling, which I object to on principle. Secondly, this will sit in a loop forever if the arming/mode change fails. This can happen because it is not possible to set modes/armed from certain states. We don't process the message that tells us if the other commands failed, so there is no way of determining what is going on.

So the question is, can/should we process command failure, and if so how?

The easiest way would probably be to make the flush() api synchronous and wait on responses to all preceding set messages (leaving exception if they fail). I wouldn't like to do it using callbacks because that is very messy code. We could just live with the code above.

@mrpollo - any ideas?

P.S. Don't know if it is possible in Python, but .NET does this quite nicely - you can make asynchronous requests and (at some point) wait on the result synchronously. This would allow us to write the code above without polling, and still return a value on failure. This is the problem with spending time in other languages, you learn about things you wish you had :-)

[ggregory8]

Hi, I agree there should be some method to identify whether a command has failed.

Another common example is arming failing. As mentioned above an easy workaround is issuing the command then polling the state until it is set, an timeout would allow you to detect this has failed but is messy also.

Are there currently any functions to handle command acknowledges?

[zlite]

Adding Fredia, the Droidplanner/Tower lead, who has been thinking about
the same thing.
On May 17, 2015 7:33 PM, "ggregory8" notifications@github.com wrote:

Hi, I agree there should be some method to identify whether a command has
failed.

Another common example is arming failing. As mentioned above an easy
workaround is issuing the command then polling the state until it is set,
an timeout would allow you to detect this has failed but is messy also.

Are there currently any functions to handle command acknowledges?

—
Reply to this email directly or view it on GitHub
#114 (comment)
.

[m4gr3d]

If the command (i.e: arm) is sent to the autopilot via the COMMAND_LONG mavlink message, as done in DroneKit-Android, then the autopilot will reply with a COMMAND_ACK message to give feedback whether the command was executed.

This mechanism can be used to detect if the command successfully reached the autopilot, and the COMMAND_ACK result field can also indicate the execution result, as described by the MAV_STATUS mavlink enum.

[hamishwillee]

Yes. I've been thinking it might be good to change our generic send_mavlink to being a message that has a callback, passing that response on return.

The problem would be to decide how you might make that work well with our current model of having settable python attributes - it might be that we need to have another model in this case.

[m4gr3d]

I'm not familiar with the python threading model, but using the described mechanism, it's possible to implement a synchronous api in java by having the executing thread block on a lock, and have it awaken by a watchdog thread when the result is confirmed, to continue execution.

However with DroneKit-Android, I decided to go with an asynchronous api as blocking on the executing thread is a no-go on mobile, and it nicely fits the exhibited client-server interaction.

[hamishwillee]

Thanks! AFAIK both models are possible on Python. Its really more a question of how we want people to be able to use the API - and move away from our dead simple but not quite effective approach. A question for @mrpollo

[AkshayBudhiraja]

Has there been any progress with regards to the direction DroneKit might be taking with this?

Command failure detection is absolutely crucial be it through a having a callback for send_mavlink or having other asynchronous calls for critical applications.

[m4gr3d]

@AkshayBudhiraja DroneKit-Android is in the process of adding the implementation discussed above, which will allow support for explicit callbacks.

[hamishwillee]

@AkshayBudhiraja Not implemented yet on DroneKit-Python. We do consider it important and will be prioritising it in the coming weeks.