Advantages and limitations of reservoir computing on model learning for robot control
Athanasios Polydoros, Lazaros Nalpantidis, Volker Krüger · Lund University Publications (Lund University) · 2015
In certain cases analytical derivation of physicsbased models of robots is difficult or even impossible. A potential workaround is the approximation of robot models fromsensor data-streams employing machine learning approaches.In this paper, the inverse dynamics models are learned byemploying a learning algorithm, introduced in [1], which isbased on reservoir computing in conjunction with self-organizedlearning and Bayesian inference. The algorithm is evaluatedand compared to other state of the art algorithms in termsof generalization ability, convergence and adaptability usingfive datasets gathered from four robots in order to investigateits pros and cons. Results show that the proposed algorithmcan adapt in real-time changes of the inverse dynamics modelsignificantly better than the other state of the art algorithms.