Reinforcement learning for bio-inspired target seeking

Abstract

Because animals are extremely effective at moving in their natural environments they represent an excellent model to implement robust robotic movement and navigation. Braitenberg vehicles are bio-inspired models of animal navigation widely used in robotics. Tuning the parameters of these vehicles to generate appropriate behaviour can be challenging and time consuming. In this paper we present a Reinforcement Learning methodology to learn the sensori-motor connection of Braitenberg vehicle 3a, a biological model of source seeking. We present simulations of different stimuli and reward functions to illustrate the feasibility of this approach.