THE LEG CONTROL OF WALKING ROBOT BY MINIMIZATION PROCESS

článek ve sborníku ve WoS nebo Scopus

en

The paper describes possible way of discretization of continuous state space of four-leg robot through simultaneous behaviour compositions, which are obtained by activation of simple basic controllers instances. The compositions are activated from non-deterministic state machine. The corresponding evaluation of state machine transitions is assigned to particular compositions of controllers in such a way, that the estimation of future profit from single transitions is sequentially improved and simultaneous compositions of controllers with highest estimate of future profit are activated hereby. The task is studied in 2-D space, corresponding to the robot platform movement paralelly to terrain plane, with respect to static stability of the robot. The behaviour obtained through three minimization methods realizing simultaneous activations of controllers was tested.

The paper describes possible way of discretization of continuous state space of four-leg robot through simultaneous behaviour compositions, which are obtained by activation of simple basic controllers instances. The compositions are activated from non-deterministic state machine. The corresponding evaluation of state machine transitions is assigned to particular compositions of controllers in such a way, that the estimation of future profit from single transitions is sequentially improved and simultaneous compositions of controllers with highest estimate of future profit are activated hereby. The task is studied in 2-D space, corresponding to the robot platform movement paralelly to terrain plane, with respect to static stability of the robot. The behaviour obtained through three minimization methods realizing simultaneous activations of controllers was tested.

mechatronics, walking robots, machine learning, Q-learning

2003

24.09.2003

Košice, SK

80-89061-77-X

Electrical Drives and Power Electronics 2003

Neuveden

Neuveden

6