Bilevel Optimization of the Helicopter Hovering Control Task (2015)
Author: Jason Zhi Liang and Risto Miikkulainen
This demo shows the performance improvement on the helicopter hovering control task when using evolved neuroevolution hyperparameters that are discovered through bilevel optimization. The goal of the helicopter hovering task is to keep the helicopter from crashing and remain as close as possible to a fixed point in space while subject to a constant wind speed. The first two videos compare the behavior of the best controller discovered through neuroevolution after 500 evaluations when using default and evolved hyperparameters. The second two videos compared default and evolved performance after 5000 evaluations.



Helicopter behavior with default neuroevolution hyperparameters after 500 evaluations. Notice that the hovering is unstable and the helicopter crashes.



Helicopter behavior with evolved neuroevolution hyperparameters after 500 evaluations. Notice that the hovering is still unstable but lasts longer.



Helicopter behavior with default neuroevolution hyperparameters after 5000 evaluations. Notice that the hovering is stable, but is unable to remain on the target in the center.



Helicopter behavior with evolved neuroevolution hyperparameters after 5000 evaluations. Notice that the hovering is stable and maintains very close distance to the target in the center.

Jason Zhi Liang Ph.D. Alumni jasonzliang [at] utexas edu
Risto Miikkulainen Faculty risto [at] cs utexas edu
Evolutionary Bilevel Optimization for Complex Control Tasks Jason Zhi Liang, Risto Miikkulainen In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO 2015), 871–878... 2015

Neuroevolution: Harnessing Creativity in AI Model Design Sebastian Risi, David Ha, Yujin Tang, Risto Miikkulainen To Appear In , Cambridge, MA, 2025. MIT Press. 2025