Viet-Hung Dang

BA*: an online complete coverage algorithm for cleaning robots

This paper presents a novel approach to solve the online complete coverage task of autonomous cleaning robots in unknown workspaces based on the boustrophedon motions and the A* search algorithm (BA*). In this approach, the robot performs a single boustrophedon motion to cover an unvisited region until it reaches a critical point. To continue covering the next unvisited region, the robot wisely detects backtracking points based on its accumulated knowledge, determines the best backtracking point as the starting point of the next boustrophedon motion, and applies an intelligent backtracking mechanism based on the proposed A* search with smoothed path on tiling so as to reach the starting point with the shortest collision-free path. The robot achieves complete coverage when no backtracking point is detected. Computer simulations and experiments in real workspaces prove that our proposed BA* is efficient for the complete coverage task of cleaning robots.

Monte-Carlo tree search for Bayesian reinforcement learning

Bayesian model-based reinforcement learning can be formulated as a partially observable Markov decision process (POMDP) to provide a principled framework for optimally balancing exploitation and exploration. Then, a POMDP solver can be used to solve the problem. If the prior distribution over the environment’s dynamics is a product of Dirichlet distributions, the POMDP’s optimal value function can be represented using a set of multivariate polynomials. Unfortunately, the size of the polynomials grows exponentially with the problem horizon. In this paper, we examine the use of an online Monte-Carlo tree search (MCTS) algorithm for large POMDPs, to solve the Bayesian reinforcement learning problem online. We will show that such an algorithm successfully searches for a near-optimal policy. In addition, we examine the use of a parameter tying method to keep the model search space small, and propose the use of nested mixture of tied models to increase robustness of the method when our prior information does not allow us to specify the structure of tied models exactly. Experiments show that the proposed methods substantially improve scalability of current Bayesian reinforcement learning methods.

Nonlinear controls of a rotating tower crane in conjunction with trolley motion

Based on two nonlinear control techniques composed of partial feedback linearization and sliding mode control, the robust nonlinear controllers are successfully designed in the case of crane’s complicated operation in which a combination of trolley translation and tower rotation is considered. The proposed controllers concurrently implement four duties well: rotating tower and moving trolley to desired positions from their initial positions precisely, keeping small the cargo swings during transport process, and completely suppressing them at cargo destination. The simulation results show the high qualities of system responses and asymptotical stability of all state trajectories. The robustness of two suggested controllers is also analyzed and compared through simulation.

Enhancing the accuracy of rainfall-induced landslide prediction along mountain roads with a GIS-based random forest classifier

Along mountain roads, rainfall-triggered landslides are typical disasters that cause significant human casualties. Thus, to establish effective mitigation measures, it would be very useful were government agencies and practicing land-use planners to have the capability to make an accurate landslide evaluation. Here, we propose a machine learning methodology for the spatial prediction of rainfall-induced landslides along mountain roads which is based on a random forest classifier (RFC) and a GIS-based dataset. The RFC is used as a supervised learning technique to generalize the classification boundary that separates the input information of ten landslide conditioning factors (slope, aspect, relief amplitude, toposhape, topographic wetness index, distance to roads, distance to rivers, lithology, distance to faults, and rainfall) into two distinctive class labels: ‘landslide’ and ‘non-landslide’. Experimental results with a cross validation process and sensitivity analysis on the RFC model parameters reveal that the proposed model achieves a superior prediction accuracy with an area under the curve  of 0.92. The RFC significantly outperforms other benchmarking methods, including discriminant analysis, logistic regression, artificial neural networks, relevance vector machines, and support vector machines. Based on our experimental outcome and comparative analysis, we strongly recommend the RFC as a very capable tool for spatial modeling of rainfall-induced landslides.

BoB: an online coverage approach for multi-robot systems

Online complete coverage is required in many applications, such as in floor cleaning, lawn mowing, mine hunting, and harvesting, and can be performed by single- or multi-robot systems. Motivated by the efficiency and robustness of multi-robot systems, this study proposes a solution to provide online complete coverage through a boustrophedon and backtracking mechanism called the BoB algorithm. This approach designs robots in the system according to a market-based approach. Without a central supervisor, the robots use only local interactions to coordinate and construct simultaneously non-overlapping regions in an incremental manner via boustrophedon motion. To achieve complete coverage, that is, the union of all covered regions in the entire accessible area of the workspace, each robot is equipped with an intelligent backtracking mechanism based on a proposed greedy A* search (GA*) to move to the closest unvisited region. The robots complete the coverage task when no more backtracking points are detected. Computer simulations show that the BoB approach is efficient in terms of the coverage rate, the length of the coverage path, and the balance of the workload distribution of robots.

Batch-Theta* for path planning to the best goal in a goal set

The development of 3D cameras and many navigation-supporting sensors has recently enabled robots to build their working maps and navigate accurately, making path planning popular not just on computer graphics, but in real environments as well. Pursuing the solutions for robot path planning, this paper presents a variant of searching method Theta* for choosing the best goal among given goals and the lowest-cost path to it, called Batch-Theta*. The novelty lies at the proposed line-of-sight checking function during the searching process and the manner that the method handles the batch of goals during one search instead of repeatedly considering a single goal or blindly doing the exhausted search. The analysis and simulations show that the proposed Batch-Theta* efficiently finds the lowest-cost path to the best goal in a given goal set under Theta*’s mechanism. Graphical Abstract

Recognizing and Tagging Vietnamese Words Based on Statistics and Word Order Patterns

In Vietnamese sentences, function words and word order patterns (WOPs) identify the semantic meaning and the grammatical word classes. We study the most popular WOPs and find out the candidates for new Vietnamese words (NVWs) based on the phrase and word segmentation algorithm [7]. The best WOPs, which are used for recognizing and tagging NVWs, are chosen based on the support and confidence concepts. These concepts are also used in examining if a word belongs to a word class.

Human Organ Classifications from Computed Tomography Images Using Deep-Convolutional Neural Network

Deep neural networks (DNNs) have recently indicated outstanding performance on image learning feature tasks while Convolutional neural networks (CNNs) have been applied for classification tasks by reducing spectral variations and the spectral correlations of the model which existed in images. In this paper, we independently approached our work as a sequence of steps. We first implemented sparse autoencoders as an unsupervised algorithm to obtain learned features in two hidden layers for the DNN model by evaluating the appropriate input features and the optimal number of hidden units, which allow us to validate basic capabilities of the dataset. Secondly, we trained a deep CNN which consisted of five main convolutional layers, followed by Rectified Linear Units (ReLUs) layers, max-pooling layers, three fully-connected layers and a final softmax probability layer, to classify the high-resolution medical images of Computed Tomography (CT) into five anatomical classes, corresponding to five organs in abdominal regions. As a result, we considerably achieved the classification accuracy of 83.74 ± 3.34% in testing. We also visualized the layer representations on CT datasets, where they indicated the state-of-the-art performance, and could hold much promise to initialize further research on computer-aided diagnosis.

A Functional Optimization Method for Continuous Domains

Smart city solutions are often formulated as adaptive optimization problems in which a cost objective function w.r.t certain constraints is optimized using off-the-shelf optimization libraries. Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is an efficient derivative-free optimization algorithm where a black-box objective function is defined on a parameter space. This modeling makes its performance strongly depends on the quality of chosen features. This paper considers modeling the input space for optimization problems in reproducing kernel Hilbert spaces (RKHS). This modeling amounts to functional optimization whose domain is a function space that enables us to optimize in a very rich function class. Our CMA-ES-RKHS framework performs black-box functional optimization in the RKHS. Adaptive representation of the function and covariance operator is achieved with sparsification techniques. We evaluate CMA-ES-RKHS on simple functional optimization problems which are motivated from many problems of smart cities.