This paper presents a design method and experiments for whole-body locomotion by a modular robot. There are two types of locomotion for modular robots: a repeating self-reconfiguration and whole-body motion such as walking or crawling. For whole-body locomotion, designing a control method is more difficult than for ordinary robots because a modular robotic system can form various configurations, each of which has many degrees of freedom. This study proposes a unified framework for automatically designing an efficient locomotion controller suitable for any module configuration. The method utilizes neural oscillators (central pattern generators, CPGs), each of which works as a distributed joint controller of each module, and a genetic algorithm to optimize the CPG network. We verified the method by software simulations and hardware experiments, in which our modular robotic system, named M-TRAN II, performed stable and effective locomotion in various configurations.
- Central pattern generator (CPG)
- Evolutionary computation
- Modular robot
- Neural oscillator network