In this paper nonlinear neural ARX model (NNARX) is used for modeling and identification of the highly nonlinear PAM-based robot arm in case variations in manipulator dynamics with intrinsic disturbance characteristics. The proposed learning approach represents a simple, yet robust, mechanism for guaranteeing finite time performance of zero learning error condition. Furthermore, offline optimization of the learning scheme configuration by a modified genetic algorithm (MGA) is implemented in advance to achieve complexity reduction and performance improvement. The proposed neural identification scheme is experimentally tested on a prototype PAM-based robot arm. The results show that the NNARX neural networks identifier inherits the advantages of the predictive modal approach, such as high speed of learning and robustness, and is used in NNARXbased control scheme to follow the actual PAM-based robot arm joint trajectories with a high accuracy.
[1]
Malur K. Sundareshan,et al.
Identification and decentralized adaptive control using dynamical neural networks with application to robotic manipulators
,
1993,
IEEE Trans. Neural Networks.
[2]
Frank L. Lewis,et al.
Neural Network Control Of Robot Manipulators And Non-Linear Systems
,
1998
.
[3]
Miomir Vukobratovic,et al.
Highly efficient robot dynamics learning by decomposed connectionist feedforward control structure
,
1995,
IEEE Trans. Syst. Man Cybern..
[4]
Kyoung Kwan Ahn,et al.
System Modeling and Identification the Two-Link Pneumatic Artificial Muscle (PAM) Manipulator Optimized with Genetic Algorithms
,
2006,
2006 SICE-ICASE International Joint Conference.