West Coast UAV

11/05/2019

It’s on

The next Outback Challenge task has been released. To take place in 2020 the next edition seems to include a second robot. One taken there by a flying one. This is going to be amazing! Not only do competitors have to fly 20 km to reach Outback Joe, injured in his barn but they have …

10/01/2019

We created our first AI, which will assist farmers with mustering, counting and assessing growth rates of their livestock.

We collected aerial livestock imagery from a farm and manually labelled a small subset of 200 photos, totaling almost 3000 livestock (sheep and cows).

We then transferred the learning from an existing deep-learning machine vision model, and re-trained it to detect the livestock from the aerial imagery.

Initial results with a very small training set, and limited processing power are very encouraging - we were able to detect and correctly classify approximately 70% of the livestock with almost no false positives.

A reinforcement learning approach can then be employed where the results from the model predictions are manually corrected and then used to further train and improve the model.

Through this process we have learnt a great deal about this tool, and how powerful it can be when used in specific applications.

02/10/2018

Here's a video summary of our mission at Dalby

29/09/2018

During the mission our support aircraft travelled over 20km to reach the remote target area. Our aircraft flew over the area several times and capture many images of the target area. From these images the red and blue target is automatically found using a script that is run on-board the aircraft, and then the location is determined. As can be seen in the image the multiple detections of the target are bunched very tightly together indicating the accuracy is very good.

27/09/2018

One of our UAV in the Dalby Herald today

It’s Jim Coyne, one of our Medical Express judges.

27/09/2018

We flew late yesterday at the challenge and head great initial launches from our catapult system and both aircraft flew off towards the target area. Unfortunately we started to see altitude and airspeed oscillations on the retrieval aircraft and lost communications after receiving a message indicating the autopilot had rebooted which would cause the aircraft to crash.

The support aircraft continued the mission and took a number of photos of Joe's farm including our target (we will post once all the teams have finished competing). On the return leg we also utilised the support aircraft imaging system to try and help locate the downed retrieval aircraft. The competition organisers also launched a quad copter to try and locate the aircraft and we co-ordinated separation and searching with them from the base area. This was a great opportunity to co-ordinate real multi-UAS operation between different operators separated by several kilometers.

We eventually received confirmation that the downed aircraft had been recovered and returned the support aircraft home for an automatic landing back at base.

The wing box area of the crashed retrieval aircraft was sufficiently intact for us to recover the micro SD card which provided hugely valuable logging information on the problem. Particular thanks to Canberra UAV for their help analysing the log which diagnosed a power supply issue on both our autopilot power supplies which caused sensor issues and eventually the reboot which led to the crash.

While a disappointing mission result this was a great opportunity to attempt the mission and extensively fly beyond visual line of sight including remote imaging of a target area over 10km from our launch site and dynamically re-planning a mission to include searching co-ordinated with other UAS during the return leg.

The West Cost UAV’s downed plane with Brett our head field Marshall.

26/09/2018

26/09/2018

It’s a hands free mission so far

24/09/2018

We've been successfully testing at Dalby before scrutineering for the competition starts tomorrow. We've run through lots of imaging to make sure we can find Joe and completed a few short full cycles of the mission including remote landing next to the target! Good progress lots of work to go!