Koldo Basterretxea

A System-on-Chip Development of a Neuro–Fuzzy Embedded Agent for Ambient-Intelligence Environments

This paper presents the development of a neuro-fuzzy agent for ambient-intelligence environments. The agent has been implemented as a system-on-chip (SoC) on a reconfigurable device, i.e., a field-programmable gate array. It is a hardware/software (HW/SW) architecture developed around a MicroBlaze processor (SW partition) and a set of parallel intellectual property cores for neuro-fuzzy modeling (HW partition). The SoC is an autonomous electronic device able to perform real-time control of the environment in a personalized and adaptive way, anticipating the desires and needs of its inhabitants. The scheme used to model the intelligent agent is a particular class of an adaptive neuro-fuzzy inference system with piecewise multilinear behavior. The main characteristics of our model are computational efficiency, scalability, and universal approximation capability. Several online experiments have been performed with data obtained in a real ubiquitous computing environment test bed. Results obtained show that the SoC is able to provide high-performance control and adaptation in a life-long mode while retaining the modeling capabilities of similar agent-based approaches implemented on larger computing machines.

Optimised PWL recursive approximation and its application to neuro-fuzzy systems

In this work, a piecewise-linear (PWL) function approximation is described by a lattice algebra. The maximum (@?) and minimum (@?) lattice operators have been modified to incorporate interpolation capability of generated PWL function vertexes. As a result of that, a new recursive method called centred recursive interpolation (CRI) based on such operators is proposed and analysed for successive function smoothing and more accurate approximation. The resultant computational scheme is accurate but simple, as few parameters are needed for function definition. The method is tested by applying it to the optimum approximation of some sample functions, and it turns out to be a natural quadratic approximation. Due to its advantageous characteristics and the properties that Gaussian-like function based neuro-fuzzy systems show, optimised approximation of programmable Gaussian functions has been studied in detail. A table of optimum parameters has been obtained for approximating the function through different design schemes. This constitutes a previous theoretical work for the future hardware implementation of function generators in neuro-fuzzy systems.

A hardware/software embedded agent for real-time control of ambient-intelligence environments

This paper presents the development of an embedded intelligent agent able to perform real-time control of ambient-intelligence environments. The system has been implemented as a system-on-programmable chip (SoPC) on a field programmable gate array (FPGA). The scheme used for realizing the intelligent agent is an adaptive neuro-fuzzy system (NFS) enhanced with a principal component analysis (PCA) pre-processor. The PCA pre-processing stage allows a reduction of the input dimensions (features) with no meaningful loss of modeling capability. As a consequence, the computational complexity of the system is significantly reduced, allowing its implementation on a single electronic device. The NFS-PCA agent has been tested with data obtained in a real ubiquitous computing environment test bed. Results obtained show that the agent is able to perform real-time control of the environment in a proactive and non-intrusive way, and also to adapt to changes of users preferences in a life-long mode.

An FPGA-based multiprocessor-architecture for intelligent environments

Abstract In this paper we propose a SoPC-based multiprocessor embedded system for controlling ambiental parameters in an Intelligent Inhabited Environment. The intelligent features are achieved by means of a Neuro-Fuzzy system which has the ability to learn from samples, reason and adapt itself to changes in the environment or in user preferences. In particular, a modified version of the well known ANFIS (Adaptive Neuro-Fuzzy Inference System) scheme is used, which allows the development of very efficient implementations. The architecture proposed here is based on two soft-core microprocessors: one microprocessor is dedicated to the learning and adaptive procedures, whereas the other is dedicated to the on-line response. This second microprocessor is endowed with 4 efficient ad hoc hardware modules intended to accelerate the neuro-fuzzy algorithms. The implementation has been carried out on a Xilinx Virtex-5 FPGA and obtained results show that a very high performance system is achieved.

Embedded high-speed Model Predictive Controller on a FPGA

Model Predictive Control (MPC) is a well established control strategy that is being used in an increasingly wider set of application areas. Unfortunately, the requirement of an on-line solution to a constrained optimization problem is an impediment for its application to fast dynamics plants. The hardware implementation of target-optimized, application-specific controllers on reconfigurable Field Programmable Gate Arrays (FPGA) can widespread the use of MPC to control systems that demand short sampling periods. This paper describes a first approach to the design and implementation of a fixed-point arithmetic model predictive controller adapted to a benchmark control problem with constraints. A FPGA chip is a suitable hardware platform to implement the tailored controller as it allows for its adaptation to other operating requirements and to different target systems while meeting real time constraints.

Brief paper: Generalized sampled-data hold functions with asymptotic zero-order hold behavior and polynomic reconstruction

We present a generalized sampled-data hold function that combines arbitrary z-domain zero-placement ability with zero-order hold behavior under constant input, thus exhibiting minimal intersample ripple by design. Our hold can be regarded as a generalization of the fractional-order hold, with a polynomial instead of a simple linear pattern, and therefore with as many tuning parameters as desired. Moreover, the polynomial approach turns out to provide a simple mechanism of control energy minimization. Among other benefits, all these features help to achieve qualitatively better perfect model referencing, because problematic sampling z-zeros or the intersample issues of the conventional generalized holds need no longer be endured.

Recursive sigmoidal neurons for adaptive accuracy neural network implementations

This paper describes an accuracy programmable sigmoidal neuron design and its hardware implementation. The “recursive neuron” can be adjusted to produce recursively more accurate and smoother piecewise linear approximations to the sigmoidal neural squashing function. This adaptive accuracy neuron, combined with a constructive training algorithm, can be used as the basic component for the implementation of self adaptive neural processing systems able to optimize power consumption and processing speeds when operating in applications with changing performance requirements and varying operational constraints.

Driving Behavior Signals and Machine Learning: A Personalized Driver Assistance System

The progressive integration of driver assistance systems (DAS) into vehicles in recent decades has contributed to improving the quality of the driving experience. Currently, there is a need for individualization of advanced DAS with the aim of improving safety, security and comfort of the driver. In particular, the need to adapt the vehicle to individual preferences and requirements of the driver is an important research focus. In this work, an individualized and non-intrusive monitoring system for real-time driver support is proposed. The kernel of the system is a driver identification module based on driving behavior signals and a high-performance machine learning technique. The scheme is suitable for the development of single-chip embedded systems. Moreover, most of the measurement units used in this research are nowadays available in commercial vehicles, so the deployment of the system can be performed with minimal additional cost. Experimental results using a reduced set of features are very encouraging. Identification rates greater than 75% are obtained for a working set of 11 drivers, 86% for five-driver groups, 88% for four-driver groups, and 90% for three-driver groups.

SCALABLE ARCHITECTURE FOR HIGH-SPEED MULTIDIMENSIONAL FUZZY INFERENCE SYSTEMS

This paper presents a scalable architecture suitable for the implementation of high-speed fuzzy inference systems on reconfigurable hardware. The main features of the proposed architecture, based on the Takagi–Sugeno inference model, are scalability, high performance, and flexibility. A scalable fuzzy inference system (FIS) must be efficient and practical when applied to complex situations, such as multidimensional problems with a large number of membership functions and a large rule base. Several current application areas of fuzzy computation require such enhanced capabilities to deal with real-time problems (e.g., robotics, automotive control, etc.). Scalability and high performance of the proposed solution have been achieved by exploiting the inherent parallelism of the inference model, while flexibility has been obtained by applying hardware/software codesign techniques to reconfigurable hardware. Last generation reconfigurable technologies, particularly field programmable gate arrays (FPGAs), make it possible to implement the whole embedded FIS (e.g., processor core, memory blocks, peripherals, and specific hardware for fuzzy inference) on a single chip with the consequent savings in size, cost, and power consumption. As a prototyping example, we implemented a complex fuzzy controller for a vehicle semi-active suspension system composed of four three-input FIS on a single FPGA of the Xilinxs Virtex 5 device family.

Efficient Algorithms for Accelerometer-Based Wearable Hand Gesture Recognition Systems

The rapid increase in the use of robotic systems in industrial and domestic environments makes it necessary the development of more natural interaction procedures. This paper presents the development of a user-specific hand Gesture Recognition System (GRS) based on the information of a single tri-axial accelerometer to recognize 7 different dynamic gestures for natural Human Machine Interaction (HMI). The aim of this paper is to analyze and compare different computational methods for feature extraction, dimensionality reduction, and vector classification in order to select the most suitable combination of signal processing stages that meets the performance requirements for a single-chip, wearable GRS system. These requirements are lag-free response, low size, and low power consumption while keeping high recognition accuracy. Experimental results show that the overall achievable accuracy can be up to 98% for Artificial Neural Network (ANN) and Extreme Learning Machine (ELM) predictors, and 99% for Support Vector Machines (SVM).