Belief-Driven Manipulator Visual Servoing for Less Controlled Environments

This paper presents the architecture of a feedforward manipulator control strategy based on a belief function that may be appropriate for less controlled environments. In this architecture, the belief about the environmental state, as described by a probability density function, is maintained by a recursive Bayesian estimation process. A likelihood is derived from each observation regardless of whether the targeted features of the environmental state have been detected or not. This provides continuously evolving information to the controller and allows an inaccurate belief to evolve into an accurate belief. Control actions are determined by maximizing objective functions using non-linear optimization. Forward models are used to transform control actions to a predicted state so that objective functions may be expressed in task space. The first set of examples numerically investigates the validity of the proposed strategy by demonstrating control in a two dimensional scenario. Then a more realistic application is presented where a robotic manipulator executes a searching and tracking task using an eye-in-hand vision sensor.     

Discriminative deep belief networks for visual data classification

Visual data classification using insufficient labeled data is a well-known hard problem. Semi-supervise learning, which attempts to exploit the unlabeled data in additional to the labeled ones, has attracted much attention in recent years. This paper proposes a novel semi-supervised classifier called discriminative deep belief networks (DDBN). DDBN utilizes a new deep architecture to integrate the abstraction ability of deep belief nets (DBN) and discriminative ability of backpropagation strategy. For unsupervised learning, DDBN inherits the advantage of DBN, which preserves the information well from high-dimensional features space to low-dimensional embedding. For supervised learning, through a well designed objective function, the backpropagation strategy directly optimizes the classification results in training dataset by refining the parameter space. Moreover, we apply DDBN to visual data classification task and observe an important fact that the learning ability of deep architecture is seriously underrated in real-world applications, especially in visual data analysis. The comparative experiments on standard datasets of different types and different scales demonstrate that the proposed algorithm outperforms both representative semi-supervised classifiers and existing deep learning techniques. For visual dataset, we can further improve the DDBN performance with much larger and deeper architecture.     

Multisensory teamwork: using a tactile or an auditory display to exchange gaze information improves performance in joint visual search

In joint tasks, adjusting to the actions of others is critical for success. For joint visual search tasks, research has shown that when search partners visually receive information about each other’s gaze, they use this information to adjust to each other’s actions, resulting in faster search performance. The present study used a visual, a tactile and an auditory display, respectively, to provide search partners with information about each other’s gaze. Results showed that search partners performed faster when the gaze information was received via a tactile or auditory display in comparison to receiving it via a visual display or receiving no gaze information. Findings demonstrate the effectiveness of tactile and auditory displays for receiving task-relevant information in joint tasks and are applicable to circumstances in which little or no visual information is available or the visual modality is already taxed with a demanding task such as air-traffic control. Practitioner Summary: The present study demonstrates that tactile and auditory displays are effective for receiving information about actions of others in joint tasks. Findings are either applicable to circumstances in which little or no visual information is available or when the visual modality is already taxed with a demanding task.     

Software Architecture Visualization: An Evaluation Framework and Its Application

In order to characterize and improve software architecture visualization practice, the paper derives and constructs a qualitative framework, with seven key areas and 31 features, for the assessment of software architecture visualization tools. The framework is derived by the application of the Goal Question Metric paradigm to information obtained from a literature survey and addresses a number of stakeholder issues. The evaluation is performed from multiple stakeholder perspectives and in various architectural contexts. Stakeholders can apply the framework to determine if a particular software architecture visualization tool is appropriate to a given task. The framework is applied in the evaluation of a collection of six software architecture visualization tools. The framework may also be used as a design template for a comprehensive software architecture visualization tool.     

Branch-and-bound task allocation with task clustering-based pruning

We propose a task allocation algorithm that aims at finding an optimal task assignment for any parallel programs on a given machine configuration. The theme of the approach is to traverse a state–space tree that enumerates all possible task assignments. The efficiency of the task allocation algorithm comes from that we apply a pruning rule on each traversed state to check whether traversal of a given sub-tree is required by taking advantage of dominance relation and task clustering heuristics. The pruning rules try to eliminate partial assignments that violate the clustering of tasks, but still keeping some optimal assignments in the future search space. In contrast to previous state–space searching methods for task allocation, the proposed pruning rules significantly reduce the time and space required to obtain an optimal assignment and lead the traversal to a near optimal assignment in a small number of states. Experimental evaluation shows that the pruning rules make the state–space searching approach feasible for practical use.     

视觉相似性计算的艺术图像自组织方法

为解决大量数字化艺术图像常规组织和管理复杂低效问题,提出一种基于图像相似性计算的自组织方法,对艺术图像提取了颜色、纹理、空间布局和SIFT等用于相似性计算的视觉特征表示,并根据艺术图像空间布局特点设计计算模型,试验了特征的聚类效果。采用多层版本近邻传播聚类（MLAP）算法为基础,对实验图像库进行层次化聚类,构建图像的层次化浏览结构。实验结果表明,该方法在艺术图像的管理和使用上都有着良好的性能。     

Effectiveness of auditory and tactile crossmodal cues in a dual-task visual and auditory scenario

Abstract

In this study, we examined how spatially informative auditory and tactile cues affected participants’ performance on a visual search task while they simultaneously performed a secondary auditory task. Visual search task performance was assessed via reaction time and accuracy. Tactile and auditory cues provided the approximate location of the visual target within the search display. The inclusion of tactile and auditory cues improved performance in comparison to the no-cue baseline conditions. In comparison to the no-cue conditions, both tactile and auditory cues resulted in faster response times in the visual search only (single task) and visual–auditory (dual-task) conditions. However, the effectiveness of auditory and tactile cueing for visual task accuracy was shown to be dependent on task-type condition. Crossmodal cueing remains a viable strategy for improving task performance without increasing attentional load within a singular sensory modality.

Practitioner Summary: Crossmodal cueing with dual-task performance has not been widely explored, yet has practical applications. We examined the effects of auditory and tactile crossmodal cues on visual search performance, with and without a secondary auditory task. Tactile cues aided visual search accuracy when also engaged in a secondary auditory task, whereas auditory cues did not.     

一种协调勘探和开采的遗传算法：收敛性及性能分析

提出了一种新的遗传算法结构。在该结构中,每一代的新种群由保留种 群、繁殖种群的随机种群三部分组成,而它们的相对数量则由不同的参数进行控制,这体现了该算法在运行过程中对搜索空间勘探和开采操作的协调和权衡。通过把该算法建模为齐次的有限Markov链,该文证明了该算法具有全局收敛性。对试验数据的分析表明,该算法能够有效协调算法对问题解空间的勘探和开采操作,因而在处理复杂问题时表现出较高的性能。     

Robust feature learning for online discriminative tracking without large-scale pre-training

Owing to the inherent lack of training data in visual tracking, recent work in deep learning-based trackers has focused on learning a generic representation offline from large-scale training data and transferring the pre-trained feature representation to a tracking task. Offline pre-training is time-consuming, and the learned generic representation may be either less discriminative for tracking specific objects or overfitted to typical tracking datasets. In this paper, we propose an online discriminative tracking method based on robust feature learning without large-scale pre-training. Specifically, we first design a PCA filter bank-based convolutional neural network (CNN) architecture to learn robust features online with a few positive and negative samples in the high-dimensional feature space. Then, we use a simple soft-thresholding method to produce sparse features that are more robust to target appearance variations. Moreover, we increase the reliability of our tracker using edge information generated from edge box proposals during the process of visual tracking. Finally, effective visual tracking results are achieved by systematically combining the tracking information and edge box-based scores in a particle filtering framework. Extensive results on the widely used online tracking benchmark (OTB-50) with 50 videos validate the robustness and effectiveness of the proposed tracker without large-scale pre-training.     

Computational models and task scheduling for parallel sparse Cholesky factorization

In this paper, a systematic and unified treatment of computational task models for parallel sparse Cholesky factorization is presented. They are classified as fine-, medium-, and large-grained graph models. In particular, a new medium-grained model based on column-oriented tasks is introduced, and it is shown to correspond structurally to the filled graph of the given sparse matrix. The task scheduling problem for the various task graphs is also discussed. A practical algorithm to schedule the column tasks of the medium-grained model for multiple processors is described. It is based on a heuristic critical path scheduling method. This will give an overall scheme for parallel sparse Cholesky factorization, appropriate for parallel machines with shared-memory architecture like the Denelcor HEP.     

A hybrid recommendation system considering visual information for predicting favorite restaurants

Restaurant recommendation is one of the most interesting recommendation problems because of its high practicality and rich context. Many works have been proposed to recommend restaurants by considering user preference, restaurant attributes, and socio-demographic behaviors. In addition to these, many customers review restaurants in blog articles where text-based subjective comments and various photos may be available. In this paper, we especially investigate the influence of visual information, i.e., photos taken by customers and put on blogs, on predicting favorite restaurants for any given user. By considering visual information as the intermediate, we will integrate two common recommendation approaches, i.e., content-based filtering and collaborative filtering, and show the effectiveness of considering visual information. More particularly, we advocate that, in addition to text information or metadata, restaurant attributes and user preference can both be represented by visual features. Heterogeneous items can thus be modeled in the same space, and thus two types of recommendation approaches can be linked. Through experiments with various settings, we verify that visual information effectively aids favorite restaurant prediction.     

Reinforcement learning of a continuous motor sequence with hidden states

Reinforcement learning is the scheme for unsupervised learning in which robots are expected to acquire behavior skills through self-explorations based on reward signals. There are some difficulties, however, in applying conventional reinforcement learning algorithms to motion control tasks of a robot because most algorithms are concerned with discrete state space and based on the assumption of complete observability of the state. Real-world environments often have partial observablility; therefore, robots have to estimate the unobservable hidden states. This paper proposes a method to solve these two problems by combining the reinforcement learning algorithm and a learning algorithm for a continuous time recurrent neural network (CTRNN). The CTRNN can learn spatio-temporal structures in a continuous time and space domain, and can preserve the contextual flow by a self-organizing appropriate internal memory structure. This enables the robot to deal with the hidden state problem. We carried out an experiment on the pendulum swing-up task without rotational speed information. As a result, this task is accomplished in several hundred trials using the proposed algorithm. In addition, it is shown that the information about the rotational speed of the pendulum, which is considered as a hidden state, is estimated and encoded on the activation of a context neuron.     

Task-oriented generation of visual sensing strategies in assemblytasks

This paper describes a method of systematically generating visual sensing strategies based on knowledge of the assembly task to be performed. Since visual sensing is usually performed with limited resources, visual sensing strategies should be planned so that only necessary information is obtained efficiently. The generation of the appropriate visual sensing strategy entails knowing what information to extract, where to get it, and how to get it. This is facilitated by the knowledge of the task, which describes what objects are involved in the operation, and how they are assembled. In the proposed method, using the task analysis based on face contact relations between objects, necessary information for the current operation is first extracted. Then, visual features to be observed are determined using the knowledge of the sensor, which describes the relationship between a visual feature and information to be obtained. Finally, feasible visual sensing strategies are evaluated based on the predicted success probability, and the best strategy is selected. Our method has been implemented using a laser range finder as the sensor. Experimental results show the feasibility of the method, and point out the importance of task-oriented evaluation of visual sensing strategies     

The influence of combined visual and tactile information on finger and eye movements during shape tracing.

Adaptation experiments in shape tracing were conducted to investigate finger and eye movements in various conditions of visual and tactile information. Maximum velocity, mean velocity, maximum acceleration and reacceleration point were calculated from finger movements. Number of eye fixations and lead time of eye fixation to finger position were calculated from eye movements. The results showed that for the finger movement the values of the indices studied were higher in the combined visual and tactile condition than in the visual only condition. The number of eye fixations decreased when subjects repeated the tracing and was more marked in the combined visual and tactile condition than in the visual only condition. The results suggest that finger movements become faster and use of vision is reduced when both visual and tactile information are given.     

Model Predictive Control for Stochastic Resource Allocation

In this paper, we consider a class of stochastic resource allocation problems where resources assigned to a task may fail probabilistically to complete assigned tasks. Failures to complete a task are observed before new resource allocations are selected. The resulting temporal resource allocation problem is a stochastic control problem, with a discrete state space and control space that grow in cardinality exponentially with the number of tasks. We modify this optimal control problem by expanding the admissible control space, and show that the resulting control problem can be solved exactly by efficient algorithms in time that grows nearly linear with the number of tasks. The approximate control problem also provides a bound on the achievable performance for the original control problem. The approximation is used as part of a model predictive control (MPC) algorithm to generate resource allocations over time in response to information on task completion status. We show in computational experiments that, for single resource class problems, the resulting MPC algorithm achieves nearly the same performance as the optimal dynamic programming algorithm while reducing computation time by over four orders of magnitude. In multiple resource class experiments involving 1000 tasks, the model predictive control performance is within 4% of the performance bound obtained by the solution of the expanded control space problem.     

Dynamic scheduling using a pheromone-based approach in multi-agent systems

The scheduling problems in factory domain applications usually involve many parallel machines, with each machine capable of processing several tasks. In most cases, changing the current machine state to another state to process a different task incurs additional material costs and time. If the overall system can maintain the expected performance, minimizing these state changes is very beneficial, and agent-based approaches inspired by the task allocation strategies of several social insects have gained increasing attention as solutions. The basic concept is based on the stimulus-threshold relation, and an individual agent determines whether it performs a given task or not based on two sets of terms, the environmental external stimuli for the task and the internal threshold values of all possible tasks. In this approach, selecting appropriate threshold values is directly related to the overall system performance, and we present a pheromone-based approach to obtain appropriate threshold values. Each agent maintains a limited, constant-sized task history queue of recently processed tasks, and the information of each agent is individually used to calculate the threshold values of tasks. Based on various experimental results, we show that the performance of the proposed method is comparable to those of other conventional methods.     

The influence of combined visual and tactile information on finger and eye movements during shape tracing

Abstract

Adaptation experiments in shape tracing were conducted to investigate finger and eye movements in various conditions of visual and tactile information. Maximum velocity, mean velocity, maximum acceleration and reacceleration point were calculated from finger movements. Number of eye fixations and lead time of eye fixation to finger position were calculated from eye movements. The results showed that for the finger movement the values of the indices studied were higher in the combined visual and tactile condition than in the visual only condition. The number of eye fixations decreased when subjects repeated the tracing and was more marked in the combined visual and tactile condition than in the visual only condition. The results suggest that finger movements become faster and use of vision is reduced when both visual and tactile information are given.