Biologically inspired modular structure to control the sit-to-stand transfer of a biped robot

In this study, a biologically inspired control structure to control the sit-to-stand (STS) transfer from a chair is developed and simulated. STS movement is consisted of two main phases. First phase of the movement is before leaving the seat (seat-off moment). In this phase seat reactions forces act on the body parts which are in contact with the seat. The second phase is after seat-off, where the only external forces acting on the body are ground reaction forces. A proper control algorithm of the STS transfer needs to consider switching between these two phases, which correspond to two different dynamical structures. The control structure developed and discussed in this work is based on the MOSAIC structure, proposed first by Wolpert and Kawato [1]. Original MOSAIC structure has a modular architecture which is based on multiple pairs of forward and inverse models of the dynamical system to be controlled, and each module is trained separately to learn one part of a given task. The number of effective modules is predetermined. We have developed a new method to train all modules simultaneously. This method is based on reinforcement and cooperative competitive learning, and the number of effective modules is determined automatically. In this study, the simulation was begun with four modules. Our results showed that only two modules out of four were selected to control the STS task. Responsibility of controlling the task was switched between the two modules around the seat-off moment. In the other words, the two modules were corresponding to the two phases of the STS.