Robot Brain Project CREST Development of Brain-Informatics Machines through Dynamical Connection of Autonomous Motion Primitives
Robot Brain Project
Nakamura Group
Asada Group
Tsuchiya Group
Ushio Group
Yoshizawa Group
Sasaki Group

Results | Ushio Group

Liapunov Design of Dynamical Information Processing and Transition Control of Motion Patterns
Toshimitsu Ushio*1
*1Osaka University

Efficient representations of primitive motions of humanoid robots are important in order to realize various complicated motions. A graphical representation called a state net is very useful for describing the transitions between basic positions [1]. In the state net, each static position is described by a node in a sensory space, and transitions between positions are described by directed arcs from a current node to the next node. So, both addition of new motions and modifications of motions are very easy. However, it is not included to represent periodic motions such as walk.

On the other hand, studies on design of dynamical systems which have a specified stable limit cycles have been done in nonlinear system theory. Green derived a constraint function which specifies the limit cycle and proposed a Lyapunov function based method for design of a desired dynamical system from the constraint function [2].

In this research, applying Green's method, we proposed a dynamics based presentation of periodic motions of humanoid robots [3]. After periodic sequence data of several primitive motions is reduced to a lower-dimensional data, we construct a nonlinear dynamical system where the reduced periodic motion is represented by a stable limit cycle. By experiment using HOAP-I, we show the effectiveness of the proposed method. Then, we introduced a hybrid state net, which is an extension of the state net, in order to present transitions between the primitive motions [4].

We also proposed a supervisory control system for motion planning of humanoid robots [5]. The proposed system is hierarchically structured into two levels. The lower level controls and monitors the robots using modular state nets. A modular state net is a state net representing motions of parts of the robots such as arms, legs, and so on, and whole body motions of the robots are represented by a combination of modular state nets for the parts. The upper level generates an optimal sequence of motions for user's requirements using timed Petri nets. A timed Petri net is used as an abstracted model of the set of all modular state nets, and using discrete event systems theory and optimal path searching algorithms, we find an optimal motion sequence. Moreover, applying a reinforcement learning method, we proposed a novel approach to the fast convergence to an optimal supervisor [6].

Using the proposed methods, we can store many primitive motions of humanoid robots which lead to realization of very complicated behavior.


[1] F. Kanehiro, M. Inaba, H. Inoue, and S. Hirai: Developmental Realization of Whole-Body Humanoid Behaviors Based on State Net Architecture Containing Error Recovery Functions, Proc. of the First IEEE-RAS International Conference on Humanoid Robots, 2000.
[2] D.N. Green: Synthesis of Systems with Periodic Solutions Satisfying V(x)=0, IEEE Transactions on Circuits and Systems, Vol.31, No.4, pp.317-326, 1984.

[3] Masakazu ADACHI, Toshimitsu USHIO, and Shigeru YAMAMOTO: Application of Lyapunov Function Based Synthesis of Nonsmooth Limit Cycles to Motion Generation for Humanoid Robots, Proc. of 2nd International Symposium of Adaptative Motion of Animals and Machines, SaP-II-4, March, 2003.

[4] Hideyuki TAKAHASHI, Keigo KOBAYASHI, and Toshimitsu USHIO: Generation of Periodic Motion of Humanoid Robots Using Hybrid State Nets, to appear at 2003 JACC, 2003.(in Japanese)

[5] Keigo KOBAYASHI, Atsushi NAKATANI, Hudeyuki TAKAHASHI, and Toshimitsu USHIO: Motion Planning for Humanoid Robots Using Timed Petri Net and Modular State Net, Proc. of 2002 IEEE International Conference on Systems, Man, and Cybernetics, pp. 334-339, October, 2002.

[6] Tatsushi YAMASAKI and Toshimitsu USHIO: Supervisory Control of Partially Observed Discrete Event Systems based on a Reinforcement Learning, Proc. of 2003 IEEE International Conference on Systems, Man, and Cybernetics, October, 2003.

Ushio Group
PDF(785KB) ]