Model-Assisted Stochastic Learning for Robotic Applications

Abstract

We present here a framework for the generation, application, and assessment of assistive models for the purpose of aiding automated robotic parameter optimization methods. Our approach represents an expansion of traditional machine learning implementations by employing models to predict the performances of input parameter sequences and then filter a potential population of inputs prior to evaluation on a physical system. We further provide a basis for numerically qualifying these models to determine whether or not they are of sufficient quality to be capable of fulfilling their predictive responsibilities. We demonstrate the effectiveness of this approach using an industrial robotic testbed on a variety of mechanical assemblies, each requiring a different strategy for completion.