Search

By introducing a biologically inspired robotic model that combines a modified feedback error learning, an unsupervised learning, and the viscoelastic actuator system in order to drive adaptive arm motions, this paper discusses the potential usefulness of a biomimetic design of robot skill. The feedback error learning is consistent with the cerebellar adaptation, the unsupervised learning, the synergy network adaptation, and the viscoelastic system of the muscles. The proposed model applies a feedforward adaptive scheme in the low dimensional control space and an adaptive synergy distribution to control redundant actuators effectively. The combination of the two adaptive control schemes is tested by controlling a two-link planar robot arm with six muscular actuators in the gravitational field. The simulation-based study demonstrates that the control scheme adapts the robot arm motions quickly and robustly to generate smooth, human-like motions.

This paper presents a robust approach to estimate the probability density function (PDF) from a sample data set. The approach is induced from entropy maximization using Renyi's quadratic entropy, and turns out to be equivalent to the support vector machines (SVM). Therefore, the approach has good properties of the support vector machines as a statistical function estimation method.