Abstract
Modular self-reconfigurable robot (MSRR) consists of certain number of modules, which is supposed to exhibit flexibility, robustness, and adaptability. However, its locomotion planning and control are proved difficult due to the changeable configuration and hyper-redundant kinematics. Furthermore, automatically generation of robot's adaptive behavior with undetermined environment and various numbers of modules is rarely investigated. This paper introduces an evolution approach to generate effective locomotion pattern and adaptive behaviors for MSRR in unknown environments. We use genetic algorithms (GA) to evolve robot's locomotion ability and gene expression programming (GEP) to evolve adaptive behaviors. The virtual evolution of a four-legged configuration is implemented upon self-developed UBotSim dynamic simulator. Both the simulation and hardware experiments show the effectiveness of this method.
