A kinematical approach for robot manipulator using a segmented tree neural networks with randomization

The approach of this paper is to deal with the problem of self and safe trajectory generation for a robot manipulator in an unstructured environment. To achieve this goal the segmented tree neural net for each link, and the randomization strategy with some cost function has been carefully presented in order to satisfy the additional constraints. In this paper we present a method of splitting the robot's degree-of-freedom from the manipulator's structural knowledge that provides multiple but a finite number of kinematic inverse solution. The strategy was based on the concept of small multilayer perceptron at each joint to extract a feature. The redundancy advantage was used to prune the possible solutions from the cost function by minimizing it, in terms of obstacle avoidance, sensor-motor torque minimization, degeneracies avoidance, and joint limit avoidance. Our method accomplished two objectives. First, it expands the power of neural tree classification using structural approaches. Second, it demonstrated the power of using random choices to minimize the specified cost function. It can said that, the neural network architecture is somehow similar to our brain architecture and randomization is often provided by our nature. Thus, the present paper can be seen as the fusion of these two inherent human natures towards the intelligent control of a robot manipulator.

Actions (login required)

View Item