A controller design method for human-assist robot with force filtering process

This paper describes a controller design method for robots supporting human in real environment. Such robots have to cope with various interactions between robots and human environment. Furthermore, they also have to autonomously classify the interaction from sensor information and select their action according to tasks and situations. This study introduces an idea of spatial filter to construct such a control system. Spatial filter is based on the idea of functionality. The idea is to decompose a large-scale control system into units called “function”. A function corresponds to an action of the control system such as “grasping”, “moving” and so on. A complicated task is achieved by superpositioning multiple functions. To construct function-based controllers, the arm coordinate space is transformed into a new coordinate space based on function modes. Fig. 1 shows composition of spatial filter. Spatial filter is an idea that treats a motion control system as a kind of a filter. It gathers information from spatially-arranged sensors. The sensor information is decomposed into function modes by a coordinate transformation matrix. Controller composition is determined based on a filter characteristic on each mode. A controller with force feedback passes force while a controller without force feedback cuts it. Controllers give input to the robot through inverse modal decomposition. Position response of the robot depends on force input from human and characteristic of spatial filter. Fig. 2 shows the overview of an experiment with spatial filter. The experimental system is composed of 3 twin-drive parallel-link manipulators. Table 1 shows functions applied then. SC, RC, GR and VC denote spring coupling, rigid coupling, grasping and velocity control function, respectively. When the operator maneuvered one of the manipulators in Step 1, all three manipulators moved only in grasping mode and accomplished open-close motion. An object was grasped in Step 2 after the operator inserted it. The object was tilted in the pitching mode in Step 3 when the operator applied force in the z direction. On the other hand, the object went up and down in Step 4 when the operator applied force in the same direction. The robot was always obedient to the operator on the mode with spring coupling function, the function with force control. On the other hand, the robot kept constant position on the mode with rigid coupling function. Rigid coupling