您好,欢迎来到华中科技大学自动化学院官网!

联系我们| English

当前位置: 首页 > 科学研究 > 学术报告 > 正文

学术报告第201836期

发布时间:2018-06-22 编辑:张惠兰 来源:科研

报告题目:Learning-based Non-linear Adaptive Robust Control

Framework: Dual-arm Robot

报 告 人:Thanana Nuchkrua

报告时间:201873日 10:00

报告地点:南一楼中311

报告摘要:

Dual-arm robot has been recognized that its functionality is attractive to meet several requirements in smart manufacturing systems, in which the automation and machining systems has integrated to be smart system. Fast response and high precision motion control of dual-arm robot are an indispensable method.

The bottlenecks of improving the control performance of the dual-arm robot are a) the parametric and nonlinear uncertainties in uenced by an unpredictable environment; b) an inherent non-linear dynamics of mechanism structure of dual-arm robot. In general, these uncertainties and non-linearity are unknown a prior and frequently change, further complicating the problem of uncertainty.

Note, in general, an application, i.e. robot arm control, of the most deterministic robust control based on xed-model approach is con gured for improved performance, but only guarantees closed-loop stability for smaller perturbations/uncertainties of the nominal system. It may not deliver the desired performance under unpredictable environments and highly non-linear uncertainty. It therefore is di cult to identify the actual dynamics of the dual-arm robot system that results in designing a non-adaptive controller to perform well under unknown and variant dynamics is not feasible.

Here we develop a parameter-based adaptive robust control framework to tackle these challenges, where a Bayesian learning-based control approach is proposed to deal with the parameter adaptation. We introduce an approach of sparse Bayesian learning (SBL), where the procedure of model selection does not require to identify parameters for adaptation law. It does not require large dimension of set values interested. In other words, it means that one can reduce the explosion of parameters if higher number of set values are employed. In addition, in terms of prediction-based, resulting in SBL also reveal that it has a good performance in prediction with the guarantee of global convergence so that we can employ SBL to predict the future dynamics of dual-arm robot. The e ectiveness of our proposed approach has been demonstrated: (i) Insensitivity to parameter uncertainties; (ii) Insensitivity to unknown payload variations; (iii) Low demand for on-line computations

 

报告人简介:

Dr Thanana has been a visiting scholar at Control group, school of Automation in Huazhong University of Science and Technology, starting from march 2018. Prior to this, he was a postdoctoral fellow at advanced institute manufacturing with high-tech innovations (AIM-HI) in National Chung Cheng University in collaboration with National Taiwan University for two years. He achieved his PhD degree with outstanding dissertation from Sirindhorn International Institute of Technology, in year of 2015. He also served various industrial manufacturing dealing with robot manipulation control. He received Master degree with excellent thesis from Asian Institute of Technology. He was a visiting student at Politecnico di Torino during Master degree studying. He obtained the recipient postgraduate award, PhD scholarship award for outstanding students, dean award for the top undergraduate students. He is an active reviewer for many international conferences and journals. 2015-, some his work published in journal of bionic engineering has been a list of the 25 most highly cited papers of journal.

His research concerns with seeking a control framework with emphasis on the model-based control for a non-linear control problem subjected to both natural and artificial systems, i.e., robotics, manufacturing, rehabilitation inspired by biology. To aim at a framework of adaptive control with robustness, the unification of control theory and learning based on optimization algorithms, where the speed of solution is considered Convex/non-convex problem, is on track of research.