A. Shmilovici, G.H. Bakir, A. Figueras, and J. Llu´s de la Rosa
Support vector machine, machine learning, multidimensional control
Examples-based controllers use historical data to evaluate local approximation models. Large data sets make it prohibitively expensive to evaluate the best control action in real time. Support vector machines (SVM) are known for their ability to identify the minimal set of data points needed to reconstruct an optimal decision surface. A successful application is presented: the simplification of a six-dimensional robotic controller. The SVM reduced the size of the data set to 5.3% of its original size while retaining 99.7% classification accuracy, thus leading the way to online implementation. The results indicate that SVM may be highly effective for the simplification of examples-based controllers.
Important Links:
Go Back
