Servers by jDrones

Microsoft is using the autonomous soaring capability in ArduPlane


(Craig Elder) #1


(Craig Elder) #2

Documentation here http://ardupilot.org/plane/docs/soaring.html


(Andre-K) #3

To bad they use patent trolls to attack Linux, Android, and open source in general…
This needs to be mentioned whenever the Microsoft is mentioned. :slight_smile:


(MagicRuB) #4

I pinged Iain Guilliard on LinkedIn.


(wkf94025) #5

The article leaves the uninformed with the impression that the innovation was all MSFT, which if they are in fact exploiting ArduSoar, is misleading at best. No mention of Sam Tabor, Tridge, or the other innovators that got them to where they are.

Kelly


(ppoirier) #6

I think we should not rely on third person information to judge on intention of the researchers.
Please read this previous blog on the same subject, you can see that they thanks the ardupilot developper team: Microsoft AI research using Mission Planner

Another important fact is that Andrey Kolobov is active here and you can read is collaboration on the sailplane wiki : http://ardupilot.org/dev/docs/soaring-sitl-with-xplane.html


(wkf94025) #7

Thanks for pointing that out. I agree, I/we should not judge MSFT acknowledgment of heavy lifting by those we preceeded them based on journalists’ take on the subject.

Kelly


(Andrey Kolobov) #8

Indeed, I wouldn’t recommend taking publications in the press too seriously as far as technical details or credit goes. As someone pointed out, we have pretty little control over what they write, despite us mentioning the connection to ArduPilot in literally every interview we’ve given. In research publications though, we will certainly give explicit credit to ArduPilot in general and Sam’s controller in particular.

On that point, @Samuel_Tabor, do you have an academic(-ish) writeup about your soaring controller that we could cite?

We’ve done lots of live experiments by flying it on real RC sailplanes and found that, while ArduSoar’s thermalling controller is outperformed by more sophisticated techniques, it still works surprisingly well given how simple it is, if the soaring radius isn’t set wildly off. The robotics community would definitely benefit from a standalone paper/writeup about it that researchers could read and cite. Do you have something like this?

If you like, we can help you write one (with you as the first author, obviously) by sharing some of our live data, generating more by flying two Radian Pros side-by-side with different algo flavors/parameter settings and/or using our in-house simulator, and/or by assisting with the writeup itself.

Alternatively, if you already have something more detailed than a blog post (currently, the only source of info on the algorithmic side of ArduSoar), everything is even easier – you could post it, e.g., on https://arxiv.org/list/cs.RO/recent and the academic community, including us, would cite it there.

Please PM me to discuss the details.


(Samuel Tabor) #9

Hi Andrey,

I don’t have a published write-up but I am very keen to collaborate on getting one out there. I’ve pinged you an email.

Also I’m sure I’m not the only one looking forward to hearing more about the more sophisticated soaring algorithms when your research is ready for publication :slightly_smiling_face:

Cheers!


(wkf94025) #10

I agree with Sam. Very interested in seeing how you have advanced the soaring IQ. I also have multiple Radians, Radian Pro’s, and plenty of Pix* FC’s to run the latest code on. I also have a complete Arduplane build environment running under Windows Subsystem for Linux / Ubuntu.

Kelly