Augmented-SVM for Gradient Observations with Application to Learning Multiple-Attractor Dynamics



In this chapter we present a new formulation that exploits the principle of support vector machine (SVM). This formulation—Augmented-SVM (A-SVM)—aims at combining gradient observations with the standard observations of function values (integer labels in classification problems and real values in regression) within a single SVM-like optimization framework. The presented formulation adds onto the existing SVM by enforcing constraints on the gradient of the classifier/regression function. The new constraints modify the original SVM dual, whose optimal solution then results in a new class of support vectors (SV). We present our approach in the light of a particular application in robotics, namely, learning a nonlinear dynamical system (DS) with multiple attractors. Nonlinear DS have been used extensively for encoding robot motions with a single attractor placed at a predefined target where the motion is required to terminate. In this chapter, instead of insisting on a single attractor, we focus on combining several such DS with distinct attractors, resulting in a multi-stable DS. While exploiting multiple attractors provides more flexibility in recovering from unseen perturbations, it also increases the complexity of the underlying learning problem. We address this problem by augmenting the standard SVM formulation with gradient-based constraints derived from the individual DS. The new SV corresponding to the gradient constraints ensure that the resulting multi-stable DS incurs minimum deviation from the original dynamics and is stable at each of the attractors within a finite region of attraction. We show, via implementations on a simulated ten degrees of freedom mobile robotic platform, that the model is capable of real-time motion generation and is able to adapt on-the-fly to perturbations.



This work was supported by EU Project First-MM (FP7/2007–2013) under grant agreement number 248258. The authors would also like to thank Prof. François Margot for his insightful comments on the technical material.


  1. 1.
    Chiang, H., Chu, C.: A systematic search method for obtaining multiple local optimal solutions of nonlinear programming problems. IEEE Trans. Circ. Syst. I Fundam. Theory Appl. 43(2), 99–109 (1996)CrossRefMathSciNetGoogle Scholar
  2. 2.
    Dixon, K., Khosla, P.: Trajectory representation using sequenced linear dynamical systems. In: Proceedings of 2004 IEEE International Conference on Robotics and Automation, 2004 (ICRA’04), vol. 4, pp. 3925–3930. IEEE (2004)Google Scholar
  3. 3.
    Ellekilde, L., Christensen, H.: Control of mobile manipulator using the dynamical systems approach. In: Proceedings of 2009 IEEE International Conference on Robotics and Automation, 2009 (ICRA’09), pp. 1370–1376. IEEE (2009)Google Scholar
  4. 4.
    Fuchs, A., Haken, H.: Pattern recognition and associative memory as dynamical processes in a synergetic system. I. Translational invariance, selective attention, and decomposition ofscenes. Biol. Cybern. 60, 17–22 (1988).
  5. 5.
    Hoffmann, H.: Target switching in curved human arm movements is predicted by changing a single control parameter. Exp. Brain Res. 208(1), 73–87 (2011)CrossRefGoogle Scholar
  6. 6.
    Jaeger, H., Lukosevicius, M., Popovici, D., Siewert, U.: Optimization and applications of echo state networks with leaky-integrator neurons. Neural Netw. 20(3), 335–352 (2007)CrossRefMATHGoogle Scholar
  7. 7.
    Khansari-Zadeh, S.M., Billard, A.: Learning stable non-linear dynamical systems with Gaussian mixture models. IEEE Trans. Robot. 27(5), 943–957 (2011). Google Scholar
  8. 8.
    Lee, J.: Dynamic gradient approaches to compute the closest unstable equilibrium point for stability region estimate and their computational limitations. IEEE Trans. Automat. Contr. 48(2), 321–324 (2003)CrossRefGoogle Scholar
  9. 9.
    Michel, A., Farrell, J.: Associative memories via artificial neural networks. IEEE Contr. Syst. Mag. 10(3), 6–17 (1990). doi:10.1109/37.55118CrossRefGoogle Scholar
  10. 10.
    Pastor, P., Hoffmann, H., Asfour, T., Schaal, S.: Learning and generalization of motor skills by learning from demonstration. In: Proceedings of 2009 IEEE International Conference on Robotics and Automation, 2009 (ICRA ’09), pp. 763–768 (2009). doi:10.1109/ROBOT.2009.5152385Google Scholar
  11. 11.
    Rasmussen, C.: Gaussian processes in machine learning. In: Advanced Lectures on Machine Learning, pp. 63–71. Springer, Berlin (2004)Google Scholar
  12. 12.
    Reimann, H., Iossifidis, I., Schöner, G.: Autonomous movement generation for manipulators with multiple simultaneous constraints using the attractor dynamics approach. In: Proceedings of 2011 IEEE International Conference on Robotics and Automation, 2011 (ICRA), pp. 5470–5477. IEEE (2011)Google Scholar
  13. 13.
    Schaal, S., Atkeson, C., Vijayakumar, S.: Scalable techniques from nonparametric statistics for real time robot learning. Appl. Intell. 17(1), 49–60 (2002)CrossRefMATHGoogle Scholar
  14. 14.
    Schölkopf, B., Smola, A.: Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond. MIT Press, Cambridge (2001)Google Scholar
  15. 15.
    Schöner, G., Dose, M.: A dynamical systems approach to task-level system integration used to plan and control autonomous vehicle motion. Robot. Auton. Syst. 10(4), 253–267 (1992)CrossRefGoogle Scholar
  16. 16.
    Schöner, G., Dose, M., Engels, C.: Dynamics of behavior: theory and applications for autonomous robot architectures. Robot. Auton. Syst. 16(2), 213–245 (1995)CrossRefGoogle Scholar
  17. 17.
    Shukla, A., Billard, A.: Coupled dynamical system based armhand grasping model for learning fast adaptation strategies. Robot. Auton. Syst. 60(3), 424–440 (2012). doi:10.1016/j.robot.2011.07.023.

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  1. 1.École Polytechnique Fédérale de Lausanne (EPFL)LausanneSwitzerland

Personalised recommendations