Intelligent Computational Scheme for Robot Inverse Kinematic Transformations

Abstract

This paper describes an intelligent computational scheme to obtain feasible solutions to the problem of inverse kinematics in robotics. The proposed scheme consists of a recurrent neural network and a knowledge-based (KB) system. The latter, based on the requirements and specifications of the robot task-space provided by the user, determines the neural network configuration and checks the robot link angles during the process of computation against the physical constraints. On the other hand, a recurrent neural network provides low-level computational features such as functional approximation, parallelism, learning and adaptation capabilities. The inverse kinematics problem in robotics involves the determination of joint variables for a desired end-effector position in the robot task-space. This problem is difficult in the sense that for a given end-effector position there can be many solutions, and some of these solutions may not be practically feasible due to the physical limitations imposed by the robot structure. The computational scheme discussed in this paper avoids time consuming analytical calculations in sharp contrast to the conventional inverse kinematics algorithms. A recurrent neural network is used in the proposed computational scheme to overcome the slow learning of the feedforward multi-layer neural networks. Furthermore, in a manner that is typical of neural networks, it would be very easy to modify the learned associations upon changes in the structure (number of links) of the robot manipulator. The KB system in the proposed scheme represents the declarative form of knowledge, while the neural network the reflexive form of information processing. Neural network and KB system are used to emulate shifts between declarative and reflexive mechanisms believed present in biological control system. The effectiveness of the proposed computational scheme is demonstrated for two- and three-linked robots.

Indexing terms:

• Neural networks
• Intelligent computation
• Robotics
• Inverse kinematics