Neuroevolutionary reinforcement learning for generalized helicopter control

By Rogier Koppejan, Shimon Whiteson. 2009

47 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Theory and Mathematics

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computational Theory and Mathematics

Citations in the corpus

Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems (2020, methodology): “Approaches have included using Gaussian process models for controlling blimps (Ko et al., 2007), and using linear regression (Koppejan and Whiteson, 2009) and locally-weighted Bayesian regression (Bagnell and Schneider, 2001) for helicopter control.”
Optimism-driven exploration for nonlinear systems (2015, methodology): “The best algorithm submitted to the 2008 Reinforcement Learning Competition for a similar, but not identical, helicopter hovering task required about 30 - 60 minutes of interaction and the use of resets, though other variants required as little as 10 minutes when provided with a good prior policy [2”
A comprehensive survey on safe reinforcement learning (2015, methodology): “While this approach has been proven advantageous in numerous applications of evolutionary methods (Hernández- Díaz et al., 2008; Koppejan and Whiteson, 2009), Koppejan's algorithm nevertheless seems somewhat ad-hoc and designed for a specialized set of environments.”
Safe Exploration of State and Action Spaces in Reinforcement Learning (2012, methodology): “In the work of Koppejan et al. (2009, 2011), neural networks are also evolved, beginning with one whose weights corresponds to teacher behavior.”
Evolutionary Computation for Reinforcement Learning (2012, methodology): “A similar approach was used to maximize on-line performance in the helicopterhovering events in recent Reinforcement Learning Competitions (Whiteson et al, 2010a): models learned via linear regression were used as fitness functions for policies evolved off-line via neuroevolution (Koppejan and White”
The Reinforcement Learning Competitions (2010, methodology): “Koppejan's agent (Koppejan and Whiteson 2009) used evolutionary computation (Goldberg 1989) to train a multilayer feed-forward neural network controller for the helicopter.”
Protecting against evaluation overfitting in empirical reinforcement learning (2011, background): “For example, consider the Generalized Helicopter Hovering environment [13], one of several generalized environments used in the recent Reinforcement Learning Competitions [14].”