An Intelligent System Approach to the Dynamic Hybrid Robot Control

The objective of this study was to solve the robot dynamic hybrid control problem using intelligent computational processes. In the course of problem- solving, biologically inspired models were used. This was because a robot can be seen as a physical intelligent system which interacts with the real world environment by means of its sensors and actuators. In the robot hybrid control method the neural networks, fuzzy logics and randomization strategies were used. To derive a complete intelligent state-of-the-art hybrid control system, several experiments were conducted in the study. Firstly an algorithm was formulated that can estimate the attracting basin boundary for a stable equilibrium point of a robot's kinematic nonlinear system. From this point the Artificial Neural Networks (ANN) based solution approach was verified for the inverse kinematics solution. Secondly, for the intelligent trajectory generation approach, the segmented tree neural networks for each link (inverse kinematics solution) and the randomness with fuzziness (coping the unstructured environment from the cost function) were used. A one-pass smoothing algorithm was used to generate a practical smooth trajectory path in near real time. Finally, for the hybrid control system the task was decomposed into several individual intelligent control agents, where the task space was split into the position-controlled subspaces, the force-controlled subspaces and the uncertain hyper plane identification subspaces. The problem was considered as a blind-tracking task by a human.

PDF FK_1996_1_A.pdf Download (791kB)

Actions (login required)

View Item