Andrew Tridgell ("Tridge") is a well-known programmer in the open source world (Samba, etc) and one of the leaders of the APM dev team. His UAV team, CanberraUAV, will be competing in this year's Outback Challenge with an APM-powered Telemaster, and at the Australian Linux conference he gave a spectacular lecture about the team's strategy and the technological challenges in the competition.
This is a must watch for anyone interested in more advanced UAV functions. Tridge and his team have added a Panda board to APM to do image processing. This is a great opportunity to watch a world-class technologist explain the strategy they're using to try to win this competition.
Comments
Just look for the car Andrew, great work.
@David, we got around 20m error at 100m altitude in our testing so far. Yes, we'd use the same attitude estimate, we're hoping that stabilising the cameras will improve things as it makes the geometry much simpler - we will be always seeing a square piece of ground, rather than the more distorted shape we get when the camera is at an angle.
We chose full automation as we thought it would be the most challenging/fun :-)
@Andrew, very well done overview of your approach to tackling the beast that is the OBC. Couple question related to the direct geo-referencing of the images: 1. What altitude is your 20m of error based on? 2. I assume you will be using the same attitude estimate from the APM to stabilize the cameras as you are now currently using for direct geo-referencing, how is this going to improve accuracy? (the more nadir the camera points, the less attitude uncertainty affect geo-referencing error, is this where you plan on seeing improvements or is there something more?). Also, very ambitious to attempt complete automation the first year out, can you highlight some reasons for this approach as opposed to mastering the minimum completion requirements first? Again, great presentation, best of luck to you and your team.
@squalish, we take short exposures to minimise the effects of vibration and movement on the cameras
We get altitude via the barometer for geo-fencing. For the OBC we don't set a geo-fencing altitude however, its a GPS fence only. For normal APM usage (eg. R/C training) altitude is used.
geo-referencing of images is via simple geometry, using GPS position plus the height and attitude from the APM telemetry stream (which is received by the pandaboard). We've found it gets a result within about 20m. We hope to improve on that once we gyro-stabilise the camera.
Why do they need 1/15000s exposures?
How is altitude determined for geofencing?
How are images geotagged and photogrammetrically referenced accurately enough for a precision drop?
Very cool stuff.
Good highlight of a mane problem with the workforce. We are each multifaceted, multilayered, complicated people, yet what do employers want? Jobs doing just 1 thing. He still just writes SAMBA drivers for a living.
Funny that he considers 2009 early on.
good work tridge, your an inspiration to us all
cool ...
Why would Tridge wanna ballistic drop something on someone? Wicked...
-
1
-
2
-
3
of 3 Next