An Auto-organization Bio-inspired Robotic System
This paper discusses and presents the possibility to endow a robotic system with some of the human capabilities (perception, data processing, recognition, learning, reasoning and acting), in order to obtain a movement behavior, with no collisions. Until now, various solutions for autonomous robots performing an obstacle avoidance task were made available in the literature. Among these different methods and techniques -ranging from the traditional logic controllers to neural controllers -, the learning neural network techniques together with the learning reinforcement paradigms are ones of the most plausible from the biologically and the in-real environment practical implementation point of view. Yet, due to the complexity of the proposed practical solutions, the corresponding integrated on-line learning process proves usually to be a slow one, and consequently, the new introduced autonomous agent concepts become often ineffective. Our new approach to the problem of obstacle avoidance task, presented in what follows, essentially differs from the previous methods through the high simplicity of the adopted solution: a) a minimal multilayer perceptron (MLP) architecture, and b) a very simple error function used with the backpropagation algorithm. The obtained improvements in the system performance are significant, and, what is more, the analytic form proposed for the error criteria allows for a further major improvement in the system feasibility.
robot neural network bio-inspired biological plausibility auto-organization
Dan-Marius Dobrea Monica-Claudia Dobrea
Electronics, Telecommunications and Information Technology Gheorghe Asachi Technical University Iasi, Romania
国际会议
2010 International Conference on Future Information Technology(2010年未来信息技术国际会议 ICFIT 2010)
长沙
英文
943-947
2010-12-14（万方平台首次上网日期，不代表论文的发表时间）