Emergence of intelligent behavior from a minimalistic stochastic model for the navigation of autonomous Robots

We use a probabilistic transition matrix methodology to realize an algorithm for the autonomous navigation of Robots. This is achieved without the necessity to set any symbolic or empiric rules, but with a learning strategy based on a purely stochastic approach. The system is tested for its abilities to exit a maze in a minimized time, results show that collisions are avoided with very high percentage of error, nearly 100%. Moreover, goals are reached in a randomly generated maze in a time range better than 80% shorter than with a non-trained algorithm. The robot it is not aware of its position nor it knows the location of the goals. The simple training with one dimensional, no memory Markovian model demonstrates the emergence of the ability to solve the maze in minimal time, a feature that we perceive as intelligent behaviour. The model is very simple to implement, does not require the definition of particular rules nor is related to a specific problem. In fact, this approach can be applied generally to any other situation where there are transitions between a finite set of internal or external states defined by sensors or actuators.