Evolving walking robots for global task based design

Abstract

The paper proposes an original method for the design of complex mechatronic systems such as locomotion systems. A modular approach is used in the hardware and software design of these robotic systems. The objective of the design process is to find the optimal combination of hardware and software components to achieve a locomotion task in a complex environment. For this we use an evolutionary algorithm, integrating a dynamic simulation of the mobile robot in its environment. A hybrid encoding allows us to evolve the robot structure and its control system simultaneously. Specialized genetic operators have been designed to manipulate this encoding and to adapt their action to the evolving population. The robot performances are evaluated through a simulation in which all criteria are computed and exploited in real time. The process is illustrated through a set of design examples. An improvement of computing time using a hierarchical evaluation is analyzed.