Team Thunder is a Sydney based team of RC enthusiasts who competed in this year's UAV Outback Search and Rescue Challenge in Kingaroy, Queensland, Australia, running the week of September 22-26, 2014. (http://www.uavoutbackchallenge.com.au/). The following is a description of our mission flight. A more technical report will follow.
The Flight
Friday came around and finally it was Team Thunder’s turn to find Joe.
The wind was picking up over 15 knots but we knew our X8 flying wing was up for it.
Setup went smooth enough until the GCS operator (Peter) discovered that the USB serial link to the RFD900 radio link wasn’t being recognised by the Windows GCS laptop. A reboot but Windows wouldn’t start up again. A little bit of panic and thinking “What will Tridge think of these fools who decided to rely on Windows for a mission critical device?” A quick change to Marks backup PC got things going again. Luckily Windows eventually started on the main GCS PC again so things were back on track.
A bit of a time out due to air traffic, then started again and we had the plane on the launcher at the end of setup time. David had all the flight systems up, Mark had the bungee loaded, Lloyd had the image recognition system up running and communicating over the 3G and RFD900 links. The GCS was all ready to go so the thumbs up was given to the pilot Tim pushed the lever on the launcher. 9:00am and Thunderbird 1 was away.
https://www.facebook.com/video.php?v=841139339252746
Before we knew it, the plane was passing EL1 and then EL2 into the search area. Smiles all round. This was exciting.
As there was a strong south-westerly blowing we had decided on north south runs, starting from the east to avoid overshooting downwind turns. We would be flying at 120m AGL, scanning 100m strips, leaving 50m gaps between the strips on the first pass that would be cleaned up on the second pass.
The first downwind turn was a little nerve racking but were soon at the north-eastern end of the search area and into our search runs.
Things were going well. Although it was windy and the plane was jumping around a bit, it was doing what it was meant to be doing and the upwind turns were working out ok.
Midway through the 3rd scan run and only 12 minutes into the mission, Mark called out that he had an automatic (thanks to Tridge’s algorithm) Joe position.
With a set of coordinates from Lloyd we got them away to the judges who passed on permission for us to drop the bottle.
We then passed over the Joe a number of times to get a more accurate averaged position. We uploaded the bottle drop waypoints. We were taking things very slowly and carefully now. We had heaps of battery and though the wind was strong, the X8 was handling it ok. Once we were happy we set off the sequence with a predefined routine from the GCS and the command “godrop” was given. Soon the bottle was away and Thunderbird 1 was on its way back to the airfield.
The marshals confirmed that the package separated from the aircraft ok to our relief.
3 minutes later, Thunderbird 1 was spotted approaching the airfield, ready to attempt an automatic landing. We underestimated the wind a little on our east-west landing at the end of the runway ending up in the long grass but a safe landing in one piece nonetheless.
https://www.facebook.com/video.php?v=841151805918166
The BOM recorded maximum wind for Friday was 43 km/h at 09:34. This was just as we landed.
The judges confirmed we had dropped the bottle within 100m of Joe and that 500ml of water had been recovered. Mission accomplished.
More details including documents, Joe Recognition System and flight logs to follow.
Team Thunder (left to right)
Tim Fu: Pilot and airframe
Lloyd Breckenridge: Joe detection system
Peter Wlodarczyk: Leader and software
Mark Frasca: Search, bottle drop and airframe
David Creusot: Avionics
Acknowledgements:
Andrew Tridgell and Canberra UAV: For the extraordinary open sourcing of everything they did, including APM software, MAVProxy, simulation environment, Joe recognition algorithm, RFD900 radio firmware, Joe image sets as well as advice during the course of the competition.
Michael Oborne: Developer of Mission Planner
The general Ardupilot developer community.
Orion Integration (http://www.orionintegration.com.au) : For the use of radios and camera and the tolerance of skunkworks by Lloyd.
Claudio Natoli: For the kind loan of a failsafe board.
Millswood Engineering (http://www.millswoodeng.com.au): For the design and kind donation of a spare failsafe board.
Seppo at RFDesign (http://rfdesign.com.au): For an awesome RFD900 radio design and advice.
Hawkesbury Model Air Sports (HMAS) Club (http://www.hmas.org.au): Support, encouragement and an excellent field at Vineyards.
Comments
Great job guys!
It was good to catch up with you at the event and trade some ideas. I liked your onboard web server to get the thumbnails down. Having the recognition running on the aircraft computer is now probably the most successful way to do SAR imaging, although on the day it was exactly what failed on our own aircraft, forcing us to use our backup cloud server and upload 1.3GB over 4G to do recog/imaging, which was just a little too demanding for good old Testra out in the sticks! BTW you can get through the Telstra NAT with the right codes applied to a corporate account that are normally used to allow VPN services through. It also gives you a semi static IP and allows it to run directly to the aircraft devices without the need of a proxy. As usual they have it but they just don't advertise it!
What surprises me is that apparently you too had the same driver issue with the RFD900 on a windows machine that we had and that forced us to swap the RFD900 out and reboot etc at the event. We have never seen it before. Did it not recognize the device in MP or in the windows system in general? We had our own frustrations (ahem) with MP as the latest versions had broken geofence and erroneous geofence messages so we used the CUAV package instead. That in turn also had some issues with uploading the geofence, which only resolved itself after reloading the PXH firmware form scratch. Many late night hours went into solving it before the event even though we had tens of hours running it on SITL at home with only minor geofence issues! For next time we'll probably have that on our own GCS as well to avoid any time released bugs and I'm very happy James got it sorted in the end as he was running out of hair to pull out! ;-)
The X8 was the airframe of choice at the event and had just the right combination of range and carrying capability to complete the challenge. Half of those who made a successful drop were flying an X8, including the first team to complete the challenge (SFWA), with a lot of other teams using very customized builds. In particular the Dutch team had, at least in my opinion, by far the best X8 build I saw at the event. It was more a work of art and aerodynamically as "perfect" as you can get a foam X8 frame. Shame there wasn't any points for design! With the right battery and propulsion combination the X8 should be able to cover around 5-7sqkm per hour, which is great for an essentially hand launched electric powered system that can be carried to site.
But in the end it's the combination of functioning systems that operate in a timely fashion that get you through to the big points, which is where your systems pulled through! Great work!
Thunderbirds are go! :-)
Thanks for the shout out Peter! It was great to be a (very miniscule) part of Team Thunder's success.
I'm well impressed at how things came together over the final few weeks - once again, congratulations to the whole team!
The camera images were recorded using network video recording software that I have developed. All images can have metadata attached in a jpeg comment. In this case I wrote a cut down version of mavproxy to run on the linux board on the aircraft.
This mavproxy like process gathers all the mavlink data needed to to be able to geolocate detections on the image and and the capture process then stores that metadata with the image as it is recorded.
There is another process which looks at every image recorded and uses CUAV's scanner algorithm to detect likely Joes. If it finds anything of interest in the image then an alarm for each detection is put into a sqlite alarm database on board the aircraft.
On the ground station we periodical query for new alarms using a web page. Cropped portions of the image for each detection are also displayed. Because we are requesting from the ground station rather than pushing images over the connection we can manage the available bandwidth for the connection type.
We had two data connections to the aircraft one via and RFD900 using pppd and another one via a via 4G to a Amazon AWS server running a proxy server that I have developed that allows for direct http connections to the board on the plane even though it is behind Telstra's NAT.
In the end the 4G connection was very good and the pppd link struggled to stay up on the FDR900 link.
We intend to do a blog post with more technical details in the near future.
Congratulations to all involved, well done. Could you guys post your parameter file for the X8?
Truly inspiring ... Congratulations!
Keen to know how you integrated Mavproxy, the image recognition system and the various automated decision making processes with Mission Planner. Looking forward to read more!
Well done guys, it was a pleasure to meet you at the event and I am glad you got to the bottle drop stage, especially after having to wait the whole week for your turn! It certainly was the year of the X8 at the outback challenge!
Thanks Tridge. Our success was substantially due to your efforts.
What I am very happy with is that we demonstrated that the Ardupilot platform is at the stage where only a modest amount of effort is required to develop and deploy a practical S&R system. I was pleasantly surprised just how well it worked in what I thought was pretty challenging conditions. The Ardupilot controlled the plane well, MAVProxy/Mission Planner allowed us to plan, configure monitor the plane the system fairly easily and your image recognition algorithm was very adaptable to our platform.
Our team member Lloyd has plans to get web browser based squinting interface that he developed for the challenge, up on a Amazon server for everyone to play around with.
There are glimpses that this kind of system could be an affordable, viable ancillary capability of professional civilian agencies such as fire brigades or emergency and rescue services and possibly even volunteer emergency organisations such as you'd find in the Australia Outback.
As for what next... at the moment we probably have a bit of family catch-up to do and after all the focus of the Outback Challenge, we've also got a fair bit of pent-up free RC plane flying to vent.
After that, I'm not sure. Even though our expertise rose greatly during the OBC I still feel we were playing catchup with you guys at Canberra UAV. Some kind of collaboration with your group and some of the other groups around Sydney may be productive.
great stuff! I was so delighted when you guys completed the challenge. An X8 is a much more practical plane for real S&R than our aircraft was, as it can be launched/landed anywhere.
Do you guys have any plans on what you might do next?