Erik Zamora

Recent advances on simultaneous localization and mapping for mobile robots

Abstract This paper gives recent advances made in simultaneous localization and mapping (SLAM) in the last decade. It summarizes the main contributions, insights, limitations, and solutions of the most popular SLAMs. Some important approaches that cannot be included in the above classifications are also discussed, such as pedestrian SLAM, SLAM with enriched maps, multi-robot SLAM, active SLAM, and SLAM in dynamic environments.

Dendrite Morphological Neural Networks trained by Differential Evolution

A new efficient training algorithm for a Dendrite Morphological Neural Network is proposed. Based on Differential Evolution, the method optimizes the number of dendrites and increases classification performance. This technique has two initialisation ways of learning parameters. The first selects all the patterns and opens a hyper-box per class with a length such that all the patterns of each class remain inside. The second generates clusters for each class by k-means++. After the initialisation, the algorithm divides each hyper-box and applies Differential Evolution to the resultant hyper-boxes to place them in the best position and the best size. Finally, the method selects the set of hyper-boxes that produced the least error from the least number. The new training method was tested with three synthetic and six real databases showing superiority over the state-of-the-art for Dendrite Morphological Neural Network training algorithms and a similar performance as well as a Multilayer Perceptron, a Support Vector Machine and a Radial Basis Network.

Ellipsoid SLAM: a novel set membership method for simultaneous localization and mapping

The extended Kalman filter (EKF) simultaneous localization and mapping (SLAM) requires the uncertainty to be Gaussian noise. This assumption can be relaxed to bounded noise by the set membership SLAM. However, the published set membership SLAMs are not suitable for large-scale and online problems. In this paper, we use ellipsoid algorithm for solving SLAM problem. The proposed ellipsoid SLAM has advantages over EKF SLAM and the other set membership SLAMs, in noise condition, online realization, and large-scale problem. By bounded ellipsoid technique, we analyze the convergence and stability of the ellipsoid SLAM. Simulation and experimental results show that the proposed ellipsoid SLAM is effective for online and large-scale problems such as Victoria Park dataset.

Differential evolution training algorithm for dendrite morphological neural networks

Abstract Dendrite morphological neural networks are emerging as an attractive alternative for pattern classification, providing competitive results with other classification methods. A key problem in the design of these neural networks is the election of the number of their dendrites. Most training methods are heuristics that do not optimize the learning parameters. Therefore, we propose a new training algorithm for classification tasks based on an optimization approach: differential evolution. We show that the besought method increases classification performance and also optimizes the number of dendrites. For generating the initial population of hyper-boxes, we adopt two techniques: one based on the division of an initial hyper-box, and the other on an initial clustering using the so-called k-means++. Both alternatives were tested on four synthetic and 11 real databases as benchmarks overcoming the state-of-the-art morphological neuron training methods as well as the radial basis networks. The proposed training algorithm achieved a favorable average accuracy compared with the well-known multilayer perceptrons and support vector machines. In addition, a real-life problem was solved by this method to recognize geometric figures using a Nao robot.

Comparing Deep and Dendrite Neural Networks: A Case Study

In this paper, a comparative study between two different neural network models is performed for a very simple type of classificaction problem in 2D. The first model is a deep neural network and the second is a dendrite morphological neuron. The metrics to be compared are: training time, classification accuracies and number of learning parameters. We also compare the decision boundaries generated by both models. The experiments show that the dendrite morphological neurons surpass the deep neural networks by a wide margin in terms of higher accuracies and a lesser number of parameters. From this, we raise the hypothesis that deep learning networks can be improved adding morphological neurons.

Dendrite morphological neurons trained by stochastic gradient descent

Dendrite morphological neurons are a type of artificial neural network that work with min and max operators instead of algebraic products. These morphological operators allow each dendrite to build a hyper-box in classification N-dimensional space. In contrast with classical perceptrons, these simple geometrical representations, hyper-boxes, allow the proposal of training methods based on heuristics without using of an optimisation method. In literature, it has been claimed that these heuristics-based trainings have advantages: there are no convergence problems, perfect classification can always be reached and training is performed in only one epoch. This paper shows that these assumed advantages come with a cost: these heuristics increase classification errors in the test set because they are not optimal and learning generalisation is poor. To solve these problems, we introduce a novel method to train dendrite morphological neurons based on stochastic gradient descent for classification tasks, using these heuristics just for initialisation of learning parameters. We add a softmax layer to the neural architecture for calculating gradients and also propose and evaluate four different methods to initialise the dendrite parameters. Experiments are performed based on several real and synthetic datasets. Results show that we can enhance the testing accuracy in comparison with solely heuristics-based training methods. This approach reaches competitive performance with respect to other popular machine learning algorithms. Our code developed in Matlab is available online.

Dendrite Ellipsoidal Neuron

A novel and efficient Dendrite Ellipsoidal Neuron based on hyper-ellipsoids is proposed. By using the clustering algorithm k-means++, the method automatically sets an optimum number of dendrites and increases classification performance. The proposed network overcomes the actual Dendrite Morphological Neural Networks due to it changes hyper-boxes by hyper-ellipsoids that create smoother decision boundaries. This technique automatically generates clusters which are converted to hyper-ellipsoids; these hyper-ellipsoids set geometric boundaries and are used to assign patterns to the corresponding classes. The new training method was tested with three synthetic and eight real databases showing superiority over the state-of-the-art for Dendrite Morphological Neural Network training algorithms and a good performance over Multilayer Perceptrons, Support Vector Machines and Radial Basis Function Networks.

Recognizing Motor Imagery Tasks Using Deep Multi-Layer Perceptrons

A brain-computer interface provides individuals with a way to control a computer. However, most of these interfaces remain mostly utilized in research laboratories due to the absence of certainty and accuracy in the proposed systems. In this work, we acquired our own dataset from seven able-bodied subjects and used Deep Multi-Layer Perceptrons to classify motor imagery encephalography signals into binary (Rest vs Imagined and Left vs Right) and ternary classes (Rest vs Left vs Right). These Deep Multi-Layer Perceptrons were fed with power spectral features computed with the Welch’s averaged modified periodogram method. The proposed architectures outperformed the accuracy achieved by the state-of-the-art for classifying motor imagery bioelectrical brain signals obtaining 88.03%, 85.92% and 79.82%, respectively, and an enhancement of 11.68% on average over the commonly used Support Vector Machines.

Morphological Neural Networks with Dendritic Processing for Pattern Classification

Morphological neural networks, in particular, those with dendritic processing (MNNDPs), have shown to be a very promising tool for pattern classification. In this chapter, we present a survey of the most recent advances concerning MNNDPs. We provide the basics of each model and training algorithm; in some cases, we present simple examples to facilitate the understanding of the material. In all cases, we compare the described models with some of the state-of-the-art counterparts to demonstrate the advantages and disadvantages. In the end, we present a summary and a series of conclusions and trends for present and further research.

Novel Autonomous Navigation Algorithms in Dynamic and Unknown Environments

Because the range of mobile robot sensor is limited and the navigation map is not accurate, autonomous navigation in dynamic and unknown environment is a big challenge. In this paper, we propose two novel autonomous navigation algorithms, which are based on the analysis of three conditions for unobserved and uncertainty environment during the navigation. The algorithm for dynamic environment uses the “known space” and “free space” conditions. It corrects false obstacles in the map when conventional path is stuck. The navigation algorithm for unknown environment uses the “unknown space” and “free space” conditions. We use Monte Carlo method to evaluate the performance of our algorithms and the other methods. Experimental results show that our autonomous navigation algorithms are better than the others.