Adaptive controllers exploiting body dynamics

Recent research has established impressive facts on how important passive body dynamics is for locomotion. Well designed bodies have interesting intrinsic dynamics, which can lead to locomotion modalities. Since the modalities are intrinsic to the body small control inputs are sufficient to elicit them.
However, so far it is not clear what good control strategies for such "interesting" bodies are. The controller needs to work together with the body, being able to elicit, stabilize and also modulate the intrinsic dynamical modes. Traditional control concepts are not well suited for this task as they are based on linear concepts and high gain control. It becomes more and more clear that a system controlling another system needs to be of the same type. I.e. for oscillatory movements a system which has intrinsic oscillatory dynamics might be a sensible choice.

Nonlinear dynamical systems are known for their pattern formation and coordination capabilities. Thus, we investigate the use of nonlinear dynamical systems for the control of robots with passive dynamics. We are especially interested in adaptive controllers which can find and adapt to the locomotion modalities of a given body and adapt to changing body properties.
We have outlined some theoretical considerations for the choice of the controller and shown implementation on examples in simulation with very promising and interesting results (cf. references). The images show the mechanical structure of a simple hopper robot (a), the feedback structure of the adaptive frequency controller (b) and the emergent bound-like gait. The gait that is generated by the adaptive controller optimizes the ratio of forward velocity and energy consumption. This is a somewhat surprising result since the system does not have explicit notion neither of forward velocity nor of energy consumption. The maximization principles that are at work need to be investigated.

The principle has been implemented on a real robot: PUPPY II (designed by F. Iida). This robot is an under-actuated quadruped robot. The knee joints are passive, but have springs, giving the robot a very pronounced body dynamics in form of resonant frequencies.


The following figure shows a readaptation experiment on PUPPY II. The weight is changed and the adaptive frequency oscillator immediately adapts to the new body property.



Furthermore, we have now a good theoretical understanding of adaptive frequency oscillators in feedback loops which make it easier to design controllers with such oscillators [1,2]. Future research aims at fusing the low level control with higher hierarchies and control and at devising methodologies to conceive adaptive controllers robots with passive dynamics.
People involved: Jonas Buchli, Auke Ijspeert

Movies

Related studend projects

Matteo de Giacomi: Locomotion Exploiting Body Dynamics (Semester Project 2006)
Giorgio Brambilla: Adaptive CPG for AIBO, (Master project 2005)

Related publications

[1]	J. Buchli, Engineering Limit Cycle Systems: Adaptive Frequency Oscillators and Applications to Adaptive Locomotion Control of Compliant Robots. PhD Thesis. EPFL, 2007. In press.
[2]	J. Buchli, F. Iida, and A.J. Ijspeert. Finding resonance: Adaptive frequency oscillators for dynamic legged locomotion. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3903-3909. IEEE, 2006. [ bib | .pdf ]
[3]	G. Brambilla, J. Buchli, and A.J. Ijspeert. Adaptive four legged locomotion control based on nonlinear dynamical systems. In From Animals to Animats 9. Proceedings of the Ninth International Conference on the Simulation of Adaptive Behavior (SAB'06) , volume 4095 of Lecture Notes in Computer Science. Springer Verlag, 2006. [ bib | http | .pdf ]
[4]	J.Buchli, L.Righetti, and A.J. Ijspeert. A dynamical systems approach to learning: a frequency-adaptive hopper robot. In Proceedings of the VIIIth European Conference on Artificial Life ECAL 2005, Lecture Notes in Artificial Intelligence, pages 210-220. Springer Verlag, 2005. [ bib | .pdf | Movie 1 | Movie 2 | Movie 3]
[5]	J.Buchli, L.Righetti, and A.J. Ijspeert. Adaptive dynamical systems for movement control. In Proceedings of 3rd International Symposium on Adaptive Motion in Animals and Machines - AMAM 2005, page7. Verlag ISLE, Ilmenau, 2005. Abstract. [ bib | .pdf ]
[6]	J. Buchli and A.J. Ijspeert. A simple, adaptive locomotion toy-system. In S. Schaal, A.J. Ijspeert, A. Billard, S. Vijayakumar, J. Hallam, and J.A. Meyer, editors, From Animals to Animats 8. Proceedings of the Eighth International Conference on the Simulation of Adaptive Behavior (SAB'04), pages 153-162. MIT Press, 2004. [ bib | .ps | .pdf ]
[7]	J. Buchli and A.J. Ijspeert. Distributed central pattern generator model for robotics application based on phase sensitivity analysis. In A.J. Ijspeert, M. Murata, and N. Wakamiya, editors, Biologically Inspired Approaches to Advanced Information Technology: First International Workshop, BioADIT 2004, volume 3141 of Lecture Notes in Computer Science, pages 333-349. Springer Verlag Berlin Heidelberg, 2004. [ bib | http | .pdf ]
[8]	L. Righetti, J. Buchli, and A.J. Ijspeert. Dynamic hebbian learning in adaptive frequency oscillators. Physica D, 216(2):269-281, 2006. [ bib | http | .pdf ]