Self-Organising Locally Interpolating Maps in Control Engineering

The work is motivated by problems during automated motion control of mobile microrobots. Microrobots are only cm-sized but can manipulate objects in the sub-micrometre-range. For a high positioning accuracy the microrobot’s actuator controller must be correspondingly accurate. But since a microrobot’s motion behaviour is difficult to model and since the model parameters even change with time, a data-based learning controller model is required. The approach of the Self-Organising Locally Interpolating Map (SOLIM) includes a new method that continuously and interpretably maps from a grid of input support vectors, e.g. a robot’s velocity, to a grid of output support vectors, e.g. corresponding control commands. Moreover, a learning algorithm has been developed, which iteratively adapts the output support vectors such that the SOLIM map represents an inverse model of the unknown system to be controlled. The most important properties of the SOLIM approach have been proven in simulations and during position control of mobile microrobot platforms.