A Control Architecture to Achieve Manipulation Task Goals for a Humanoid Robot

Focusing on the manipulation tasks to be executed by humanoid robots, principal requirements which are to be satisfied by hardware/software of the control system are considered. In order to meet the requirements, a novel type of hardware structure and software architecture is proposed in this paper. Since the target humanoid robot consists of multiple subsystems such as a central controller for brain, a vision controller for eye, and five motion sub-controllers for two arms, two hands, one spine, the on-board hardware control system is designed to have a distributed control structure connected by pseudo realtime Ethernet interfaces. A goal-achieving software architecture is also proposed which meets the requirements of semi-autonomy, reactivity, expandability, and object-orientedness. Specifically, in order to achieve the reactivity, a coordination method is proposed to configure three kinds of executive modules, primitive module, flow-control module, and goal module, which have multiple exit states. The control architecture proposed has been implemented for performing toy-block assembly tasks on a humanoid robot as well as on the graphic simulator.