Estimation and Mapping of System-Surface Interaction by Combining Nonlinear Optimization and Machine Learning

摘要

Exploring unknown environments is one of the main applications of mobile robotic systems. Since explorative trajectories can be used to gather information on the environment as well as on the internal dynamics of the robotic system, we propose a combined parameter estimation and mapping approach consisting of three steps: first, the parameter estimation problem is addressed by nonlinear optimization. Then, clustering is used to classify the estimated parameters. Finally, Support Vector Machines (SVMs) are used to expand the optimal parameter values of the recorded data onto the entire map. The proposed approach is applied to a wheeled mobile rover system in a scenario with sharply changing surface properties. Further, it is assumed that the model structure is known and the slip parameter is estimated during the exploration as it depends on system-surface interaction. From the simulations, it was demonstrated that the proposed approach can estimate the position-dependent slip parameter and identify more than 90% of the surface map.