Antonio M. López

An Incremental Adaptive Life Long Learning Approach for Type-2 Fuzzy Embedded Agents in Ambient Intelligent Environments

In this paper, we present a novel type-2 fuzzy systems based adaptive architecture for agents embedded in ambient intelligent environments (AIEs). Type-2 fuzzy systems are able to handle the different sources of uncertainty and imprecision encountered in AIEs to give a very good response. The presented agent architecture uses a one pass method to learn in a nonintrusive manner the users particular behaviors and preferences for controlling the AIE. The agent learns the users behavior by learning his particular rules and interval type-2 Membership Functions (MFs), these rules and MFs can then be adapted online incrementally in a lifelong learning mode to suit the changing environmental conditions and user preferences. We will show that the type-2 agents generated by our one pass learning technique outperforms those generated by genetic algorithms (GAs). We will present unique experiments carried out by different users over the course of the year in the Essex Intelligent Dormitory (iDorm), which is a real AIE test bed. We will show how the type-2 agents learnt and adapted to the occupants behavior whilst handling the encountered short term and long term uncertainties to give a very good performance that outperformed the type-1 agents while using smaller rule bases

Real-time gait event detection for normal subjects from lower trunk accelerations

In this paper we report on a novel algorithm for the real-time detection and timing of initial (IC) and final contact (FC) gait events. We process the vertical and antero-posterior accelerations registered at the lower trunk (L3 vertebra). The algorithm is based on a set of heuristic rules extracted from a set of 1719 steps. An independent experiment was conducted to compare the results of our algorithms with those obtained from a Digimax force platform. The results show small deviations from times of occurrence of events recorded from the platform (13+/-35 ms for IC and 9+/-54 ms for FC). Results for the FC timing are especially relevant in this field, as no previous work has addressed its temporal location through the processing of lower trunk accelerations. The delay in the real-time detection of the IC is 117+/-39 ms and 34+/-72 ms for the FC, improving previously reported results for real-time detection of events from lower trunk accelerations.

Comparison of Step Length Estimators from Weareable Accelerometer Devices

Wearable accelerometry provides easily portable systems that supply real-time data adequate for gait analysis. When they do not provide direct measurement of a spatio-temporal parameter of interest, such as step length, it has to be estimated with a mathematical model from indirect sensor measurements. In this work we are concerned with the accelerometry-based estimation of the step length in straight line human walking. We compare five step length estimators. Measurements were taken from a group of four adult men, adding up a total of 800 m per individual of walking data. Also modifications to these estimators are proposed, based on biomechanical considerations. Results show that this modifications lead to improvements of interest over previous methods

Multisensor Approach to Walking Distance Estimation with Foot Inertial Sensing

Walking distance estimation is an important issue in areas such as gait analysis, sport training or pedestrian localization. A natural location for portable inertial sensors for gait monitoring is to attach them to the user shoes. Step length can be computed by means of a biaxial accelerometer and a gyroscope on the sagital plane. But estimations based on the direct signal integration are prone to error. This paper shows the results achieved by using a multisensor model approach to reduce uncertainty. Unbounded growth of error is reduced by means of sensor fusion techniques. The method has been tested, and early experimental results show that it provides an estimation of the walking distance with a standard deviation smaller than with single IMU similar systems.

Pedestrian navigation based on a waist-worn inertial sensor.

We present a waist-worn personal navigation system based on inertial measurement units. The device makes use of the human bipedal pattern to reduce position errors. We describe improved algorithms, based on detailed description of the heel strike biomechanics and its translation to accelerations of the body waist to estimate the periods of zero velocity, the step length, and the heading estimation. The experimental results show that we are able to support pedestrian navigation with the high-resolution positioning required for most applications.

Modified Pendulum Model for Mean Step Length Estimation

Step length estimation is an important issue in areas such as gait analysis, sport training or pedestrian localization. It has been shown that the mean step length can be computed by means of a triaxial accelerometer placed near the center of gravity of the human body. Estimations based on the inverted pendulum model are prone to underestimate the step length, and must be corrected by calibration. In this paper we present a modified pendulum model in which all the parameters correspond to anthropometric data of the individual. The method has been tested with a set of volunteers, both males and females. Experimental results show that this method provides an unbiased estimation of the actual displacement with a standard deviation lower than 2.1%.

An evolutionary algorithm for the off-line data driven generation of fuzzy controllers for intelligent buildings

Ambient intelligence is nowadays an active research field. As a key matter of this concept, several approaches have been proposed for the development of learning architectures for the control of the devices in an intelligent building. In this paper, an evolutionary algorithm is analyzed as a candidate for the initial phases of the design of such architectures: fuzzy controllers for the devices are offline induced from data sampled from the environment. We would show results obtained using real data gathered from the Essex intelligent dormitory. The proposed algorithm seems to be suited for the task, both due to its accuracy and for the easy and meaningful linguistic interpretation of the solutions it produces

Ambulatory estimation of mean step length during unconstrained walking by means of COG accelerometry.

It has been reported that spatio-temporal gait parameters can be estimated using an accelerometer to calculate the vertical displacement of the bodys centre of gravity. This method has the potential to produce realistic ambulatory estimations of those parameters during unconstrained walking. In this work, we want to evaluate the crude estimations of mean step length so obtained, for their possible application in the construction of an ambulatory walking distance measurement device. Two methods have been tested with a set of volunteers in 20 m excursions. Experimental results show that estimations of walking distance can be obtained with sufficient accuracy and precision for most practical applications (errors of 3.66 ± 6.24 and 0.96 ± 5.55%), the main difficulty being inter-individual variability (biggest deviations of 19.70 and 15.09% for each estimator). Also, the results indicate that an inverted pendulum model for the displacement during the single stance phase, and a constant displacement per step during double stance, constitute a valid model for the travelled distance with no need of further adjustments. It allows us to explain the main part of the erroneous distance estimations in different subjects as caused by fundamental limitations of the simple inverted pendulum approach.

Pedestrian dead reckoning with waist-worn inertial sensors

We present a waist-worn personal navigation system based on inertial measurement units. The device makes use of the human bipedal pattern to reduce position error. We describe improved algorithms, based on detailed description of the heel strike biomechanics and its translation to accelerations of the body waits, to estimate the periods of zero velocity, the step length, and the heading estimation. The experimental results show that we are able to support pedestrian navigation with the high-resolution positioning required for most applications.

Application of techniques of dimension reduction to predict the steel quality at the end of the secondary steelmaking

Nowadays, there is not any model capable to predict the steel quality before the end of the secondary steelmaking where corrective actions are still possible. In this paper it is presented a modelization by SOM (self organizing map) of a Secondary steelmaking facility able to estimate the steel carbon content at the end of the treatment considering process variables. Different network configurations were tested for two different input variables sets in order to obtain the best result, which is measured as the minimum error committed in the estimation.