Can Isik

A trigonometric trajectory generator for robotic arms

Interpolation of a robot joint trajectory is realized using trigonometric splines. This method is based on the assumption that joint-space knots have been generated from cartesian knots by an inverse kinematics algorithm. The use of trigonometric splines results in smooth joint trajectories and is amenable to real-time obstacle avoidance. Some of the spline parameters can be chosen to minimize an objective function (e.g. jerk or energy). If the objective function is minimum jerk, a closed-form solution is obtained. This paper introduces a trajectory interpolation algorithm, discusses a method for trajectory optimization, and includes simulation examples.

Partially connected feedforward neural networks structured by input types

This paper proposes a new method to model partially connected feedforward neural networks (PCFNNs) from the identified input type (IT) which refers to whether each input is coupled with or uncoupled from other inputs in generating output. The identification is done by analyzing input sensitivity changes as amplifying the magnitude of inputs. The sensitivity changes of the uncoupled inputs are not correlated with the variation on any other input, while those of the coupled inputs are correlated with the variation on any one of the coupled inputs. According to the identified ITs, a PCFNN can be structured. Each uncoupled input does not share the neurons in the hidden layer with other inputs in order to contribute to output in an independent manner, while the coupled inputs share the neurons with one another. After deriving the mathematical input sensitivity analysis for each IT, several experiments, as well as a real example (blood pressure (BP) estimation), are described to demonstrate how well our method works.

Constrained fuzzy logic approximation for indoor comfort and energy optimization

Indoor environmental satisfaction has been receiving considerable attention by many researchers recently. Research has indicated that allowing building occupants to adjust their local environment to their liking increases satisfaction and human performance. However, concern about the possible increase of energy consumption associated with the wide adoption of distributed localized environmental control has limited the use of such systems. In this study, we show how gradient-based optimization can be used to minimize energy consumption of distributed environmental control systems without increasing occupant thermal dissatisfaction. Fuzzy rules have been generated by data from gradient optimization, showing that a fuzzy logic control scheme based on nearest neighbors approximates closely the gradient-based optimized results.

Inference engines for fuzzy rule-based control

Abstract The knowledge-based control of autonomous vehicles allows efficient hierarchical structures that utilize linguistic sensory data at various levels of resolution and exactness. This is mainly due to the fact that the control is based on a collection of rules rather than an analytical controller. Each rule in the controller prescribes the control for a specific situation. The applicability of a rule in an observed situation involves inexactness, which is modeled using fuzzy sets. The control rules can be obtained analytically, experimentally, or from an expert. All of these approaches involve certainty levels of possible control commands, and the rule bases can best be represented as fuzzy relations. The experimental identification of a mobile robot behavior is described in this paper as a two-step process. These steps are the determination of the vocabulary of representation and the derivation of fuzzy control rules. The experiments and the derived rules are geared towards minimum-time control of the robot motion. The combination of uncertainties that exist in the rules and observations gives rise to an inference mechanism based on the extension principle. Although computationally straightforward, the sequential max and min operations involved in the inferencing are too time-consuming and may prohibit real-time operation. In this paper two architectures for the parallel computation of max-min operations are described and their applicabilities to rule-based control are compared.

Stability and convergence of neurologic model based robotic controllers

The authors investigate the local convergence properties of an artificial-neural-network (ANN)-based learning controller, using linearization techniques. The controller utilizes generic multilayer ANNs to adaptively approximate the manipulator dynamics over a specified region of the state space for a given desired trajectory. This generic neural network structure can be viewed as a nonlinear extension of a deterministic autoregressive model which is commonly used in model matching problems for linear systems.<<ETX>>

Blood pressure estimation using neural networks

Oscillometry is an indirect method to determine blood pressure. An inflatable and debatable cuff is placed on arm to observe oscillations at different pressure levels. Thus, an envelope obtained from the oscillations is related to the blood pressure. In our work, we extract few features from the oscillometric waveforms, and estimate blood pressure using feedforward neural networks. Feature strength is evaluated by computing the standard deviation of the errors. The results are compared with the traditional maximum amplitude pressure algorithm. A large noninvasively collected database is used for this purpose.

Feature subset selection for blood pressure classification using orthogonal forward selection

Systolic and diastolic pressures are used to define cardiac related health in general medicine. In our present work, we investigate blood pressure classification based on pressure waveforms using a relatively large non-invasively collected database. Feature selection is required to reduce redundant features in the data set for a better classification. Therefore, a selection method based on an orthogonal transform is used.

Adaptive fuzzy control via modification of linguistic variables

A fuzzy knowledge-based control system developed for mobile robot motion control is introduced. The method for deriving a fuzzy knowledge-base is established, based on a commercially available mobile robot. The adaptation method for fuzzy knowledge-based systems reacts to environmental variations and modifies the definitions of linguistic state variables instead of modifying the knowledge-base itself. Some simulation results are presented to demonstrate that the proposed adaptive scheme makes significant improvements in system performance even when there are severe environmental variations.<<ETX>>

Decision making at a level of a hierarchical control for unmanned robot

An Intelligent Mobile Autonomous System (IMAS), which is equipped with vision and low level sensors to cope with unknown obstacles, is modeled as a hierarchy of decision making for motion planning and control. The world description is based on linguistic variables that assume values from possible interval corresponds to the membership grade to a fuzzy set. The decision making is presented as a production system with a fuzzy database. The choice of optimal motion execution commands are performed using fuzzy set operators. Also included in the paper is the procedure of rule derivation and examples based on low level motion control.

A maximum entropy radial basis function network based neuro-fuzzy controller

This paper presents a systematic approach to constructing a self-organizing fuzzy controller. The proposed controller is built on a neuro-fuzzy system consisting of a maximum entropy self-organizing net (MESON) and a radial basis function network (RBFN). We develop the corresponding self-organizing algorithms. MESON, a new fuzzy clustering neural network model, combines the ideas of fuzzy membership values for learning rates based on the maximum entropy principle, and the structure and update rules of the Kohonen clustering network (KCN). The strategy proposed in our approach for the update rules of KCN is derived from the fixed-point iteration for the solution of nonlinear equations. This model eliminates the sensitivity to the choice of the initial configuration and yields a dynamic fuzzy clustering solution. MESON is used for the generation of fuzzy rules as well as the construction of RBFN for fuzzy inference.