Automatic Resonance Tuning and Feedforward Learning of Biped Walking using Adaptive Oscillators

In this contribution, we propose a novel approach to achieve two fundamental features of adaptive control in rhythmic movements: The automatic locking of the controller to the system’s resonance frequency, and the progressive transfer from feedback to feedforward control. This is realized by using adaptive oscillators, i.e. mathematical tools capable of synchronizing to a periodic input and learning this input’s main features (frequency, amplitude). In particular, these concepts are illustrated here on a simulated walking task. Automatic resonance tuning emerges from the interaction between the plant (the biped walker), the controller, and the movement phase directly estimated by the adaptive oscillator. Progressive transfer to feedforward control is realized by learning the control pattern (initially provided by the feedback controller) with local filters. In conclusion, this paper illustrates to potential relevance of using adaptive oscillators as central elements in adaptive control of rhythmic tasks