Modified Elman neural network with dynamic learning rate for tracking and motion prediction of a nonholonomic three-wheeled mobile robot

This article proposes to track and predict the trajectory of a non-holonomic three-wheeled mobile robot using a modified Elman neural network. An algorithm for calculating the learning rate of a neural network is suggested, which improves the efficiency and speed of learning and also reduces the number of iterations required for learning. The modified Elman algorithm with dynamic learning rate (MENN) is compared with the classical Elman neural network (ENN) and the PSO algorithm (PSO-ENN). Training of a neural network is evaluated according to two criteria: the number of iterations required for training and the average training time. In addition, the deviation from the given trajectory of movement is checked: movement along a straight line, in a square and in a circle for each algorithm. The simulation results showed that the modified Elman neural network with a dynamic learning rate is more efficient (by 32.4% on average) and accomplishes the learning objective faster (by 66.4% on average) and has the least deviation from the given motion trajectory. The relative measurement error ranges from 7.8% to 20.2% at 95% reliability and five tests for each group of measurements.

1. Su K.H., Chen Y.Y., Su S.F. Design of neural-fuzzy-based controller for two autonomously driven wheeled robot. Neurocompting. 2015; 73:2478–2488.

2. Berezina V.A., Mezentseva O.S., Ganshin K.Y. Hybrid neuroevolution as a way to train neural networks by the example to solve the maze problem. Modeling, Optimization and Information Technology. 2021;9(3). Available by: https://moitvivt.ru/ru/journal/pdf?id=1012. DOI: 10.26102/2310-6018/2021.34.3.014 (accessed 26.04.2022). (In Russ.)

3. Lucas R., Oliveira R. M., Nascimento C. B., Kaster M. S. Performance analysis of an adaptive Gaussian nonlinear PID control applied to a step-down CC-CC converter. IEEE 24th International Symposium on Industrial Electronics (ISIE). 2015;1:1022–1026.

4. Li Z., Yang C., Su C., Deng J., Zhang W. Vision-Based Model Predictive Control for Steering of a Nonholonomic Mobile Robot. IEEE Transactions on Control Systems Technology. 2019;24(2):553–564.

5. Suprapto B.Y., Mustaqim A., Wahab W., Kusumoputro B. Modified elman recurrent neural network for attitude and altitude control of heavy-lift hexacopter. 15th International Conference on Quality in Research (QiR): International Symposium on Electrical and Computer Engineering. 2017;1:309–314.

6. He W. Adaptive neural network control of an uncertain robot with fullstate constraints. IEEE Transactions on Cybernetics.2016;46(3):620–629.

7. Al-Jamali N. A., Al-Raweshidy H. S. Modified Elman Spike Neural Network for Identification and Control of Dynamic System. IEEE Access, vol. 2020;8:61246–61254.

8. Li D., Liu Y., Tong S., Chen C. L. Approximation-Based Adaptive Neural Tracking Control of Nonlinear MIMO Unknown Time-Varying Delay Systems with Full State Constraints. IEEE Transactions on Cybernetics. 2017;47(10):3100–3109.

9. Deng L., Seethaler R. J., Chen Y., Yang P., Cheng Q. Modified Elman neural network based neural adaptive inverse control of rate-dependent hysteresis. International Joint Conference on Neural Networks (IJCNN). 2016;1:2366–2373.

10. Wang S., Zhang Y., Dong Z., Du S., Ji G. Feed-forward neural network optimized by hybridization of PSO and ABC for abnormal brain detection. International Journal of Imaging Systems & Technology. 2015;25(2):153–164.