Parameter estimation of manipulator robot dynamics

Abstract

We need better methods of dynamic parameter estimation because robots are becoming more complex and their interactions with the unmodeled world are becoming more frequent. These methods should be fast, reliable, and easily adaptable to various classes of robots. Here, we design and build a parameter estimation system in a robotic simulation and controls framework called Drake. The parameter estimation works by minimizing the error between a model and the observed data from a given trajectory. Using our estimator, we then compare the minimization of three types of errors, generated by the dynamics model, energetic model, and simulation model. We quantify the speed and performance of each algorithm on the Acrobot system.