Combined biological paradigms: A neural, genetics-based autonomous systems strategy

Abstract This paper introduces an autonomous systems strategy that combines two biological inspirations: neural networks and gentic algorithms (GAs). These ideas have been combined in a variety of ways in other systems, but the scheme presented here has several unique features. The system presented is based on an analogy between learning classifier systems (LCSs) and neural networks first presented by Smith and Cribbs [Evolutionary Computation 2(1) (1994) 19-36]. However, Smith and Cribbs focused on supervised learning. The work presented in this paper transfers these ideas to the realm of autonomous systems by considering reinforcement learning. In the new system, a neural network is used to map environmental states to Q values. The neural network structure is based on an LCS. The GA acts to shape neural connectivity, and the number of hidden layer nodes. The GAs action is similar to its action in the LCS. The suggested system is evaluated in a simulated mobile robot test environment. Experimental results suggest that the system is effective in learning and evolving parsimonious strategy representations for autonomous systems. Future directions for investigation of this system are discussed.