Model Reference Learning Control using ANFIS (Masters Thesis)

abstract

A model-reference learning controller using a tuned ANFIS (Adaptive Network based Fuzzy Inference System) corrector for a nonlinear plant in general and a 2-R robot in particular. The controller uses nominal model of the plant which is likely to have model/parametric uncertainties as a reference model. A model based controller is designed based on the nominal/reference model. An ANFIS corrector adds to the output of the model based controller a correction so as to make the plant follow the behavior of the reference model. ANFIS learns the correction input using the difference between the output of the reference plant and the actual plant, which is defined as the error.

Multi-agent Search (Doctoral Thesis)

abstract

We address the problem of multi-agent  search in an unknown environment as an extension and application of
the concepts available in the literature. Autonomous agents equipped with necessary sensors carry out the search operation in the search space where the uncertainty or the lack of information is known  a priori as a probability density function. Two search strategies are proposed namely, deploy and search and continuous search. We haveshown that in each step of the deploy and search strategy, the deployment of agents is optimal in the sense of maximizing the per step reduction in the uncertainty. We prove, that the proposed strategies can reduce the uncertainty density to arbitrarily low level under ideal conditions. We provide a few formal analysis results such as stability and convergence of
the proposed control laws, and spatial distributedness of the strategies under realistic constraints such as limit on maximum speed of agents, agents moving with constant speed and limitation on sensor range. The simulation experiments show that the proposed strategies perform fairly well, and the continuous search strategy leads to somewhat shorter and smother trajectories than those of the deploy and search strategy.