Overview

• Real-Coded Extended Compact Genetic Algorithm (rECGA) [1]

• Recurrent Neural Network

• 2-D Pole Balancing Environment [2]

• Finless Rocket Environment [3]

• If time permits / future work

rECGA

Represent joint probability distributions as mixtures of Gaussians

Real-Coded Extended Compact Genetic Algorithm parameter fitting

Recurrent Neural Network

• Already implemented (many times over)

2-D Pole Balancing Environment

• Very simple environment to implement

• Source code available everywhere

Pole balancing cart

Finless Rocket Environment

• Uses open source JSBSim flight dynamics library

• Rocket definition file not available (need to contact Eric Gullichsen of Interobital Systems)

Finless rocket guidance environment

What else?

• Use linkage model to find reusable modules

• Explore alternative algorithms for model search

• Try out Real-Coded Bayesian Optimization Algorithm (rBOA) [4]

Bibliography

[1] P.L. Lanzi, L. Nichetti, K. Sastry, D. Voltini, and D.E. Goldberg, “Real-Coded Extended Compact Genetic Algorithm Based on Mixtures of Models,” Studies in Computational Intelligence, vol. 157, 2008, pp. 335–358.

[2] F. Gomez and R. Miikkulainen, “2-D pole balancing with recurrent evolutionary networks,” ICANN, 1998, p. 758–763.

[3] F. Gomez and R. Miikkulainen, “Active guidance for a finless rocket using neuroevolution,” GECCO, 2003, p. 2084–2095.

[4] C. Ahn, R. Ramakrishna, and D. Goldberg, “Real-Coded Bayesian Optimization Algorithm: …,” Lecture Notes in Computer Science, 2004, pp. 840–851.

What did I just say?

• Real-Coded Extended Compact Genetic Algorithm (rECGA)

• Recurrent Neural Network

• 2-D Pole Balancing Environment

• Finless Rocket Environment

• If time permits / future work