Evolving a Helicopter's Heading Controller

The PID controller of an autonomous helicopter robot is evolved using a genetic algorithm and the fitness is evaluated on the robot itself.

Over 30 generations, the control parameters of a PID heading controller of a small indoor helicopter evolve towards more suitable and robust solutions. To evaluate the fitness of each individual, its control parameters are used on the actual flying robot. The helicopter is artificially perturbed by 90° and -90° and the controllers ability to reach and keep the setpoint at 0° determines the fitness.

For more information please read the full paper:
B.N.Passow, M.A.Gongora, S.Coupland, A.A.Hopgood,
Real-time Evolution of an Embedded Controller for an Autonomous Helicopter,
Proc. of the IEEE Congress on Evolutionary Computation, WCCI'08, Hong Kong, 2008
A pre-print version is available here: http://www.cci.dmu.ac.uk/preprintPDF/MarioWCCI2008_CEC.pdf

Done at the Centre for Computational Intelligence, De Montfort University, Leicester, UK:
http://www.cci.dmu.ac.uk
This research is sponsored by and done in cooperation with the IOCT, Institute of Creative Technologies:
http://www.ioct.dmu.ac.uk

Category:

Education

Tags:

• helicopter
• control
• genetic
• algorithm
• GA
• PID
• robot
• AI
• artificial
• computational
• intelligence
• computer
• science
• engineering
• Flyper

[AtheistVendetta]

That's pretty slick... would be interesting to evolve the software virtually to a decent level where it can control the helicopter on its own. This has great real-world applications in surveillance UAV's and such.

[TheCCI]

@AtheistVendetta

Thanks. We evolved two of the four controllers so far and the helicopter is hovering as stable now as I can fly it manually.

I believe this is very useful for any kind of control problem where no formal model is known...for any kind of unmanned aerial vehicle (UAV) or general industrial control problem.

[TheCCI]

Actually, this setup was used to evolve (using a genetic algorithm) control parameters. We have shown that this method was successfully finding robust control parameters, automatically and without any human intervention. We tested the parameters in actual test flights and published the findings in scientific papers. Take a look at this test flight video.

[NebunLaCap]

what do i have to major in to be doing this kind of stuff in school?

[mdivinc]

What a waste of time. You can't check anything with that set up. your only fooling yourself