So it looks like it’s time for some more OBC development!
From what I can tell so far the main differences to last years event are:
- 3 separate prizes depending on level of achievement
- $25,000 for completing the mission with the highest points
- $25,000 for using type 2 autonomy (now being redefined as "no controller/user interference, but with constant comms link to controller)
- $25,000 to avoid Dynamic No Fly Zones (details yet to be disclosed)
- Manual hand launches are no longer considered autonomous (excluding them from getting type 2 prize)
- Layout of course is different, with only the first two waypoints needing to be completed in order and the inclusion of no fly zones
Still the same:
- Range and VTOL required to pickup the blood sample from the remote landing site
- Aircraft size allowances, propulsion etc
- Geofence requirements (addition of no fly zones)
- Comms required when retrieval aircraft is on the ground at remote landing site (ADS-B transponders might be introduced)
Overall it seems like the biggest change in comparison to the last one in 2016 is the emphasis on, and the extra prizes for automation. The aircraft platforms can be the same, so our last one should work fine.
Link to OBC site: https://uavchallenge.org/medical-express/
Link to rules: https://drive.google.com/file/d/0B5JgqjkRDqw8SHkyUUU1NDZTaDQ/view
Very interesting, thanks for the update.
One more good reason for us to get the stay-out-zone (aka no-fly-zone) feature working in master.
Similar to this, I’m expecting the “Japan Innovation Challenge” to post it’s rules for this year in the next couple of months. It’s a similar competition although it only held it’s first event last year. I’ll post details when they’re made public.
I’ve had a good read of the rules and a quick look at how it would be best to implement the no-fly-zones in APM.
Currently there are a few missing pieces to achieve the level of automation required:
- Would be to include static, predefined no-fly-zones inside the geofence area. This could also be achieved with fixed mission planning to avoid no-fly areas, however, the interaction with dynamic no-fly areas could become an issue if the UAV needs to avoid a dynamic no-fly zone or geofence in flight as well. A way to predefine them would be good along with the geofences, and possibly a way to priorities the direction of avoidance.
- Include dynamic no-fly zones. Currently there are no details on how the coordinates for dynamic no-flyzones will be provided, but they do talk about adding ADS-B transponders to the event. I’d expect that the coordinates would be provided via some sort of data stream to the GCS/aircraft. This would seem to indicate that there will be some similarity with ADS-B, and as such making the existing ADS-B module work within these requirements would probably be the best way to integrate it into APM. I note that ADS-B currently supports avoidance, but I am currently unaware of how it reacts to both geofence, static no-fly-zones and possibly moving and size changing dynamic no-fly-zones.
- Further to this these aircraft are typically “range challenged”. This means that plotting a course to avoid any obstacles could severely impact the available range to complete the mission. So a way to autonomously plot course corrections, maybe even considering prevailing wind conditions, might be necessary. Note for the type 2 autonomy extra prize money there has to be “no hands” interfering with the control of the aircraft after the command for launch. Also depending on when and where the dynamic no-fly-zone comes by, and how far along the aircraft is in the mission (only OBC WP1-2 need to be done it order) will directly depend on which path is the shortest to complete the mission.
I think there are also a bunch of landing approach and LIDAR optimizations that will need to be done that will need to be tightly integrated into the Joe and landing zone location recognition regimes, that need to take wind direction etc into account.
It will be interesting to see how the Japanese competition compares and if there are any challenge overlaps.
Ok, very interesting.
I can definitely imagine the ADS-B logic will need to be changed. I guess we can imagine that there’s some other aircraft with an ADS-B tranmitter that could fly through the area as our vehicle(s) are flying to Joe. Currently ArduPilot (both plane and copter) treat the arrival of such a vehicle as a failsafe event. So it stops what it was doing and tries to fly away from the other vehicle. In Copter at least, if a fence is set-up then it will stop at the fence - so it’ll try to run away but won’t run too far. Once the other vehicle is out of range (the range is configurable) then it can be configured to resume the previous flight mode (i.e. back into AUTO or Guided).
For the dynamic no-fly zones, Copter’s in pretty good shape. We just need to add some code to allow uploading and storing the zones. All the other stuff is much like the fence so it won’t be hard. I’m less sure about what changes Plane will need.
By the way, the Japan Innovation Challenge competition last year was similar to the original OBC. The main objective was to find a Joe like dummy placed within a 3km radius area. The area itself was quite hilly and forrested making it difficult to spot Joe by eye. The Pilots were stationed right at the edge of the area. There was no requirement for it to be autonomous which meant most of the competitors were using DJI equipment and running about staring at little screens try to find the dummy. A good number of vehicles were lost. It was a 5 day event and two teams found Joe. The prize this year (I hear) is 50k for finding Joe, 200k for retrieving him but it hasn’t been officially announced yet.
Even though mulit’s have been getting better, including hybrid gas/electric ones, I think that copter still won’t be feasible for running in the OBC because of the range and speed required. Ideally the avoidance action would be a navigation event, not a failsafe event in the case of the OBC. Hovering in a position will probably not be viable in the OBC due to the time constraint of completing the mission in one hour, and diminishing range reserves trying to hover a VTOL/quadplane whilst waiting for a no-flyzone to pass. In fact it’s possible that an encounter with the dynamic no-fly-zone might result in mission termination due to running out of juice, especially on airframes that are “optimized” for the range/speed and not for hover. So a predictive type of avoidance would be preferred I think, with the best resulting course plotted depending on the variables of weather and all no-fly-zones/geofences. Apart from the first two WP’s it seems we have some flexibility in where we fly when on the course, so we should have enough room to move around obstacles.
Funnily enough when I first started with quadplane development, before either PX4 or Arduplane supported it “ages” ago in 2013, I actually ran Arducopter on a PXH and strapped a pusher prop to the back of a mini hex and raced at 100kmh around the sky! I used a separate RC channel to control the pusher thrust. It worked a treat! The interesting thing was that even without wings on it at the time (Skyfun wings were added later) the overall current didn’t change much between hover and forward flight with that setup (maybe 10-20% more).
This was due to a higher pitch forward prop pushing a flat profile hex horizontally through the air instead of on angle, which meant the hex props would autorotate in forward flight resulting in less hover power required to keep it airborne. At first controlling altitude was a bit of a pain until I switched to alt hold, which basically turned the thing into an high altitude hovercraft with it drifting around the sky on the pusher prop motor and only using yaw to control direction. You’d just flick the yaw in the intended direction and use the pusher throttle to control the rate of turn, and just drift around. Using roll and pitch for a roll type turn always ended up with the thing changing altitude and hard to control. Was fun! (I have to get it out again) Afterwards with the wings there was a only a slight difference in current draw, but the thing would always climb with throttle as it had too much lift from the autorotating props and wing. We’re working a variation of this for the next event, that is basically a 50g blob of custom made avionics with a few 18650 batteries held up by a minimal wing, propelled by three motors with asymmetric lift for hover and only using two servo’s, all in a hybrid tailsitter/quadplane config under 2kg.
In fact it might be a contender for the Japanese competition provided you can return to swap batteries or fly more than one aircraft. 3km radius is a big area to search in detail with undulating terrain. Do you have a link to previous Japanese Challenge pls? Also when you say retrieve Joe, do you mean physically lift him up with a drone and bring him back to base? Now that is a challenge!