Pole-Balancing with Different Evolved Neurocontrollers

The paper presents various evolved neurocontrollers for the pole-balancing problem with good benchmark performance. They are small neural networks with recurrent connectivity. The applied evolutionary algorithm, which is not based on genetic algorithms, was designed to evolve neural networks with arbitrary connectivity. It uses no quantization of inputs, outputs or internal parameters, and sets no constraints on the number of neurons. Network topology and parameters like weights and bias terms are developed simultaneously.

[1]  Richard S. Sutton,et al.  Neuronlike adaptive elements that can solve difficult learning control problems , 1983, IEEE Transactions on Systems, Man, and Cybernetics.

[2]  J. Sitte,et al.  A cartpole experiment benchmark for trainable controllers , 1993, IEEE Control Systems.

[3]  Jürgen Moser,et al.  Dynamical Systems, Theory and Applications , 1975 .

[4]  C.W. Anderson,et al.  Learning to control an inverted pendulum using neural networks , 1989, IEEE Control Systems Magazine.

[5]  Frank Pasemann,et al.  Evolved Neurocontrollers for Pole-Balancing , 1997, IWANN.

[6]  Xin Yao,et al.  A review of evolutionary artificial neural networks , 1993, Int. J. Intell. Syst..

[7]  J. D. Schaffer,et al.  Combinations of genetic algorithms and neural networks: a survey of the state of the art , 1992, [Proceedings] COGANN-92: International Workshop on Combinations of Genetic Algorithms and Neural Networks.