Vector-valued function estimation by grammatical evolution for autonomous robot control

An autonomous mobile robot requires a robust onboard controller that makes intelligent responses in dynamic environments. Current solutions tend to lead to unnecessarily complex solutions that only work in niche environments. Evolutionary techniques such as genetic programming (GP) can successfully be used to automatically program the controller, minimizing the limitations arising from explicit or implicit human design criteria, based on the robot’s experience of the world. Grammatical evolution (GE) is a recent evolutionary algorithm that has been applied to various problems, particularly those for which GP has performed well, with additional advantages such as memory efficiency. We formulate robot control as vector-valued function estimation and present a novel generative grammar for vector-valued functions. A consideration of the crossover operator leads us to propose a design criterion for the application of GE to vector-valued function estimation, along with a second novel generative grammar which meets this criterion. The suitability of these grammars for vector-valued function estimation is assessed empirically on a simulated task for the Khepera robo