Model Calibration

Abstract

Any model, regardless of its nature, is characterized by parameters, i.e., unknown quantities that need to be determined in order to solve the model. Often, parameters are directly measurable on the target system (e.g., the weight, the size, or the speed of a robot), they can be obtained from data sheets or computed from physical laws (e.g., friction and restitution coefficients, inertia matrices, etc.). However, more abstract models might involve so-called free parameters that result from the lumping of several phenomena, some of which might be difficult to model explicitly, thereby rendering direct approaches inapplicable. In such cases, more sophisticated numerical methods and machine learning techniques are necessary.