Fully adjustable multilayer topological neural networks for intelligent autonomous system design

A neural network system is proposed to execute tasks for which cognitive autonomy is essential. The system acquires knowledge while interacting with the environment to efficiently reach its aims. No supervising process is necessary. The design exploits the animal conditioning theory to give support to the neural network reinforcement learning. The architecture consists of three main modules: a conditioned (multilayer and topological) network, an instinctive behavioral network, and a regulatory network. Any synapse of the multilayer neural network can be adjusted during learning. An autonomous control application provides an opportunity to appraise its potentialities. Simulation results confirm that the system learns how to change the environment in order to accomplish efficiently the task.