Aerobatic A.I turns on the competition . This research was published by "Flair Polo" Media
In Scientific Aviation Monthly Journal.
Written by ; Philip John-Robertson.One of the many ways researchers are working to improve the performance of autonomous aircraft is by having them compete against humans in the physical business of flying them.
It takes some skill to pilot a small aircraft accurately especially in a confined flight envelopes, and by developing computional algorithms that outperform humans, may usher in a generation of autonomous systems with incredible capabilities.
In the space of 25 years, Control line aerobatics went from an underground hobby for aircraft enthusiasts to a professional sport using the latest electric systems and composite carbon fibre materials. Among the bodies facilitating these high levels of competition, in 2019 the organizers included for the first time a dedicated competition for developers of autonomous stunt planes, who could pit their self-piloted aircraft against eachother for a significant cash prize if they can complete the full control line pattern unassisted.
Using a novel algorithm that can plot a flight path with great efficiency, scientists at the University of Zurich are now claiming to have done just that.
The C/L AFD1 developed at Delft University in Sweden claimed first prize in the inaugural event, proving 260 percent higher score than the next best autonomous stunt plane in the field moving from 140 to almost 500 in a virtual point scoring system.
Now fast forward to August 2021 in the space of less than two years, the University of Zurich researchers claim to have perfected the system and set new standards, albeit in a very different setting and with a few caveats. They say that previous algorithms for autonomous control line stunt systems would usually rely on simplifications of either the control system or the flight path itself.
Their novel a.i algorithm improves on both these by more accurately considering the limits of the plane, and calculating "time-optimal trajectories" that accelerate and decelerate the electric system at just the right rates through different segments the pattern.
“
The novelty of our algorithm is that it is the first of its kind to generate time-optimal trajectories that fully consider the physical 3D space the plane is achieving during the stunt pattern also calculating the 3D path limitations including a non consistent gravity vector,” says study lead engineer Dr Sarmuzzha.Scientists have now used this autonomous stunt plane to beat 5 of the worlds best pilots from around the world in a special invite only 7 round flyoff tournament for the first time.
The team proved the worth of their new algorithm by using it to navigate a commerically available european electric competition plane through the full aerobatic pattern.
8 external cameras were used to capture the motion of the plane and give real-time information on its location, which then informs the algorithm to calculate flight vectors, ground proximity, airspeed and 3D spatial recognition. Control of the stunt plane was then turned over to a central controller system that branched off three independent peer to peer systems running flaps and elevators with the A.I also controlling the electric motor using highly modified onboard commercially available accelerometers but with sensitivity speeds at 0.36% variance of real time adjustments and time-optimal change speed rates under 0.15 microseconds.
The unit ran on a central swivel pole attached to a robotic "control arm system" which converted the algorithmic code into physical servo movements in the centre of the circle which provided "real time" rotational movement at the handle.
All the manouvers completed by the algorithm were better statistically than all the human pilots it flew against and the performance was more consistent, as once it determined the optimal path through the pattern, with pullout heights adjusted it was able to reliably repeat all shapes with 98.5% rulebook perfection accuracy and reflected score efficiency with less than 2% error from a perfect flight across all 7 competition flights.
The scientists say this is the first time an autonomous control line stunt plane has outperformed human pilots in any stunt competition.
P.J Rowland from Australia who first heard about the system in 2019 and flew against it in the invite only 2021 tournament said ..
"It was a great experience, frightening at the same time seeing a robotic arm fly and doing those rulebook corners, best flights ive ever seen anywhere on the planet. All of us who were invited to compete in this competition agreed it was just outstanding, we also saw a demonstration of it flying just as perfectly in 65 mph winds it was sensational at speed control " Another top USA pilot said :
"
I dont think it will be that long before those of us in the Top 5 competing for the US Nats wins, might end up losing to the computer like we all did today ,atleast the A.I cant carry the plane to the circle or build its own plane, but even without those appearance points most of us would loose."
A European pilot who was involved in the original airframe build said
"
Most of us in Europe fly this style of plane but noone has ever come close to this level of rulebook flying. We may employ this system for training our juniors in the future and welcome this new pilot into our competitionsResearchers explain competition is difficult, because not only does the algorithm rely on the 8 external cameras to gauge its position within the patten, it takes about an hour of computing for it to calculate the time optimal trajectory for each pattern then another hour for the code compilation; taking up to 2 hours for it to learn the subtle nature of the entire pattern for a given day; wind direction, humidity and altitude, using a GPU and huge cloud storage to execute and run the code for each flight.
These are two factors the researchers are looking to address before the algorithm finds its way into commercial use or competition systems : reducing the computational demands of the algorithm and enabling it to rely on an onboard " brain esc " and real time anaysis.
But the algorithm still represents a significant step forward for the technology, and could prove useful for all sorts of applications from fishing, survival inspections after earthquakes to archery.
Whether they're completing search and rescue operations, inspecting damaged buildings or dominating the L-pad and FAI world championships the goal is to show mastery of all aviation domains and have them doing so with ease, speed, efficiency and reliability.
Unofficial sources at both the A.M.A and F.A.I have rumored to be considering granting an application for a temporary license to allow the a.i algorithm to compete as a sentient competitor at the 2026 US Nationals and World championships, claiming this could revolutionise the contol line stunt scene with a guaranteed 100 % probability of a first place victory , explaining that all humans pilots will be made redundant within 10 years and ushering in a new era of an A.I only stunt events.
This research was originally published by "Flair Polo" Media , in Scientific Aviation Monthly Journal
Photo of the prototype arm used by the team.
[/url]
Topic: New technology (Read 1858 times)