Automated Path Search and Optimization of Robotic Motion Using Hybrid ART-SOM Neural Networks
This paper proposes a new type of unsupervised, path optimizing artificial neural network (ANN) suitable for autonomous motion control of multi-joint robotic mechanisms with arbitrary degrees of freedom (DoF). The ANN can search through the robot’s workspace and select an optimal path avoiding obstacles, among several possible paths. This approach does not require computation of nonlinear inverse kinematic expressions generally used for such mechanisms. The proposed ANN combines features of adaptive resonance theory (ART) and self-organizing maps (SOMs). It is a sparse hierarchical multilayer deep learning network with specific features implemented at each layer. Cells in lower levels classify input using a ART/SOM hybrid structure. Higher levels will successively identify and optimize paths that can be used to solve motion problems. The paper describes the cellular automata required to implement the path optimizing network. These ANNs have been implemented using R simulation language. Results for various types of joint systems are presented.
KeywordsAdaptive resonance theory (ART) Artificial neural networks (ANNs) Deep learning Motion control Self-organizing maps (SOMs)
The authors would like to thank Siddaganga Institute of Technology, Tumakuru, C-Quad, Belagavi and KLE Dr. M S Sheshgiri College of Engineering & Technology, Belagavi for all the support.
- 1.Aguero, C.E., Koenig, N., Chen, I., Boyer, H., et al.: Inside the virtual robotics challenge: simulating real-time robotic disaster response. IEEE Trans. Autom. Sci. Eng. 12(2) (2015)Google Scholar
- 2.Wali, A.: Clojure for Machine Learning. Packt Publishing, Birmingham UK (2014)Google Scholar
- 3.Kumar, S.: Neural Networks: A classroom approach. Tata McGraw-Hill Publishing Co Ltd., New Delhi (2004)Google Scholar
- 4.Tai L., Liu M.: Deep-learning in mobile robotics—from perception to control systems: a survey on why and why not. Online: Cornell University Archives, arXiv:1612.07139v3 [cs.RO], 1 (2017)
- 6.Ai, J., Funt, B., Shi L.: A new type of ART2 architecture and application to color image segmentation, pp. 89–98. Springer, LNCS 5163, ICANN (2008)Google Scholar
- 9.Faigl, J.: An application of self-organizing map for multirobot multigoal path planning with minmax objective. Comput. Intell. Neurosci. Hindawi, 2016, Article ID 2720630. Available online http://dx.doi.org/10.1155/2016/2720630 (2016)
- 10.Sun, C., He, W., Ge, W., Chang, C.: Adaptive neural network control of Biped robots. IEEE Trans. Syst. Man Cybern. Syst. 47(2), 315–326 (2017)Google Scholar
- 11.Benbrahmin, H.: Biped dynamic walking using reinforcement learning, Ph.D. Dissertation, University of New Hampshire (1996)Google Scholar
- 12.Chetlur, S., Woolley, C.: cuDNN: efficient primitives for deep learning. Cornell University archives, arXiv:1410.0759v3 [cs.NE] (2014)
- 14.Potlonjak, V., Svetozarevic, B., Jovanovic, K., Holland, O.: The puller-follower control of compliant and non-compliant antagonistic tendon drives in robotic systems. Int. J. Adv. Rob. Syst. 8(5), 143–155 (2011)Google Scholar
- 15.Aparanji, V.M., Wali, U.V., Aparna, R.: A novel neural network structure for motion control in joints. In: ICEECCOT Mysore, Dec 2016, pp. 227–232 (2017) (also available from IEEE Xplore)Google Scholar
- 16.Aparanji, V.M., Wali, U.V., Aparna, R.: Robotic motion control using machine learning techniques. In: 6th IEEE International Conference on Communication and Signal Processing, Melmaravattur, (ICCSP 2017) Apr. 2017, IEEE Xplore (in press)Google Scholar