Olac Fuentes

Experimental evaluation of uncalibrated visual servoing for precision manipulation

We present an experimental evaluation of adaptive and non-adaptive visual servoing in 3, 6 and 12 degrees of freedom (DOF), comparing it to traditional joint feedback control. While the purpose of experiments in most other work has been to show that the particular algorithm presented indeed also works in practice, we do not focus on the algorithm but rather on properties important to visual servoing in general. Our main results are: positioning of a 6 axis PUMA 762 arm is up to 5 times more precise under visual control than under joint control; positioning of a Utah/MIT dextrous hand is better under visual control than under joint control by a factor of 2; and a trust-region-based adaptive visual feedback controller is very robust. For m tracked visual features the algorithm can successfully estimate online the m/spl times/3 (m/spl ges/3) image Jacobian (J) without any prior information, while carrying out a 3 DOF manipulation task. For 6 and higher DOF manipulation, a rough initial estimate of J is beneficial. We also verified that redundant visual information is valuable. Errors due to imprecise tracking and goal specification were reduced as the number of visual features, m, was increased. Furthermore highly redundant systems allow us to detect outliers in the feature vector and deal with partial occlusion.

Generic algorithms: what fitness scaling is optimal?

Genetic algorithms are among the most promising optimization techniques. They are based on the following reasonable idea. Suppose that we want to maximize an objective function J(x). We somehow choose the first generation of “individuals” x1, x2, [tdot],xn (i.e., possible values of x) and compute the “fitness” J(xi) of all these individuals. To each individual xi we assign a survival probability pi that is proportional to its fitness. In order to get the next generation we then repeat the following procedure k times: take two individuals at random (i.e., xi with probability pi) and “combine” them according to some rule. For each individual of this new generation, we also compute its fitness (and survival probability), “combine” them to get die third generation, etc. Under certain reasonable conditions, the value of the objective function increases from generation to generation and converges to a maximal value The performance of genetic algorithms can be essentially improved if we MX fitness scaling, i.e.,...

Object detection using image reconstruction with PCA

In this paper, we present an object detection system and its application to pedestrian detection in still images, without assuming any a priori knowledge about the image. The system works as follows: in a first stage a classifier examines each location in the image at different scales. Then in a second stage the system tries to eliminate false detections based on heuristics. The classifier is based on the idea that Principal Component Analysis (PCA) can compress optimally only the kind of images that were used to compute the principal components (PCs), and that any other kind of images will not be compressed well using a few components. Thus the classifier performs separately the PCA from the positive examples and from the negative examples; when it needs to classify a new pattern it projects it into both sets of PCs and compares the reconstructions, assigning the example to the class with the smallest reconstruction error. The system is able to detect frontal and rear views of pedestrians, and usually can also detect side views of pedestrians despite not being trained for this task. Comparisons with other pedestrian detection systems show that our system has better performance in positive detection and in false detection rate. Additionally, we show that the performance of the system can be further improved by combining the classifier based on PCA reconstruction with a conventional classifier using a Support Vector Machine.

Color-based road sign detection and tracking

This paper describes a general framework for the detection and tracking of traffic and road signs from image sequences using only color information. The approach consists of two independent parts. In the first we use a set of Gaussian distributions that model each color for detecting road and traffic signs. In the second part we track the targets detected in the first step over time. Our approach is tested using image sequences with high clutter that contain targets with the presence of rotation and partial occlusion. Experimental results show that the proposed system detects on average 97% of the targets in the scene in near real-time with an average of 2 false detections per sequence.

Harnessing the power of big data: infusing the scientific method with machine learning to transform ecology

Most efforts to harness the power of big data for ecology and environmental sciences focus on data and metadata sharing, standardization, and accuracy. However, many scientists have not accepted the data deluge as an integral part of their research because the current scientific method is not scalable to large, complex datasets. Here, we explain how integrating a data-intensive, machine learning approach with a hypothesis-driven, mechanistic approach can lead to a novel knowledge, learning, analysis system (KLAS) for discovery and problem solving. Machine learning leads to more efficient, user-friendly analytics as the streams of data increase while hypothesis-driven decisions lead to the strategic design of experiments to fill knowledge gaps and to elucidate mechanisms. KLAS will transform ecology and environmental sciences by shortening the time lag between individual discoveries and leaps in knowledge by the scientific community, and will lead to paradigm shifts predicated on open access data and analytics in a machine learning environment.

Hierarchical Learning of Navigational Behaviors in anAutonomous Robot using a Predictive Sparse DistributedMemory

We describe a general framework for learning perception-based navigational behaviors in autonomous mobile robots. A hierarchical behavior-based decomposition of the control architecture is used to facilitate efficient modular learning. Lower level reactive behaviors such as collision detection and obstacle avoidance are learned using a stochastic hill-climbing method while higher level goal-directed navigation is achieved using a self-organizing sparse distributed memory. The memory is initially trained by teleoperating the robot on a small number of paths within a given domain of interest. During training, the vectors in the sensory space as well as the motor space are continually adapted using a form of competitive learning to yield basis vectors that efficiently span the sensorimotor space. After training, the robot navigates from arbitrary locations to a desired goal location using motor output vectors computed by a saliency-based weighted averaging scheme. The pervasive problem of perceptual aliasing in finite-order Markovian environments is handled by allowing both current as well as the set of immediately preceding perceptual inputs to predict the motor output vector for the current time instant. We describe experimental and simulation results obtained using a mobile robot equipped with bump sensors, photosensors and infrared receivers, navigating within an enclosed obstacle-ridden arena. The results indicate that the method performs successfully in a number of navigational tasks exhibiting varying degrees of perceptual aliasing.

Multi-Pose Face Detection with Asymmetric Haar Features

In this paper we present a system for multi-pose face detection. Our system presents three main contributions. First, we introduce the use of asymmetric Haar features. Asymmetric Haar features provide a rich feature space, which allows to build classifiers that are accurate and much simpler than those obtained with other features. The second contribution is the use of a genetic algorithm to search efficiently in the extremely large parameter space of potential features. Using this genetic algorithm, we generate a feature set that allows to exploit the expressive advantage of asymmetric Haar features and is small enough to permit exhaustive evaluation. The third contribution is the application of a skin color-segmentation scheme to reduce the search space. Our system uses specialized detectors in different face poses that are built using AdaBoost and the C4.5 rule induction algorithm. Experimental results using the CMU profile test set and BioID frontal faces test set, in addition to our own multi-pose face test set, show that our system is competitive with other systems presented recently in the literature.

Automatic classification of pathological gait patterns using ground reaction forces and machine learning algorithms

An automated gait classification method is developed in this study, which can be applied to analysis and to classify pathological gait patterns using 3D ground reaction force (GRFs) data. The study involved the discrimination of gait patterns of healthy, cerebral palsy (CP) and multiple sclerosis subjects. The acquired 3D GRFs data were categorized into three groups. Two different algorithms were used to extract the gait features; the GRFs parameters and the discrete wavelet transform (DWT), respectively. Nearest neighbor classifier (NNC) and artificial neural networks (ANN) were also investigated for the classification of gait features in this study. Furthermore, different feature sets were formed using a combination of the 3D GRFs components (mediolateral, anterioposterior, and vertical) and their various impacts on the acquired results were evaluated. The best leave-one-out (LOO) classification accuracy 85% was achieved. The results showed some improvement through the application of a features selection algorithm based on M-shaped value of vertical force and the statistical test ANOVA of mediolateral and anterioposterior forces. The optimal feature set of six features enhanced the accuracy to 95%. This work can provide an automated gait classification tool that may be useful to the clinician in the diagnosis and identification of pathological gait impairments.

Acquiring Visual-Motor Models for Precision Manipulation with Robot Hands

Dextrous high degree of freedom (DOF) robotic hands provide versatile motions for fine manipulation of potentially very different objects. However, fine manipulation of an object grasped by a multifinger hand is much more complex than if the object is rigidly attached to a robot arm. Creating an accurate model is difficult if not impossible. We instead propose a combination of two techniques: the use of an approximate estimated motor model, based on the grasp tetrahedron acquired when grasping an object, and the use of visual feedback to achieve accurate fine manipulation. We present a novel active vision based algorithm for visual servoing, capable of learning the manipulator kinematics and camera calibration online while executing a manipulation task. The approach differs from previous work in that a full, coupled image Jacobian is estimated online without prior models, and that a trust region control method is used, improving stability and convergence. We present an extensive experimental evaluation of visual model acquisition and visual servoing in 3, 4 and 6 DOF.

A distributed evolutionary method to design scheduling policies for volunteer computing

Volunteer Computing (VC) is a paradigm that uses idle cycles from computing resources donated by volunteers and connected through the Internet to compute large-scale, loosely-coupled simulations. A big challenge in VC projects is the scheduling of work-units across heterogeneous, volatile, and error-prone computers. The design of effective scheduling policies for VC projects involves subjective and time-demanding tuning that is driven by the knowledge of the project designer. VC projects are in need of a faster and project-independent method to automate the scheduling design. To automatically generate a scheduling policy, we must explore the extremely large space of syntactically valid policies. Given the size of this search space, exhaustive search is not feasible. Thus in this paper we propose to solve the problem using an evolutionary method to automatically generate a set of scheduling policies that are project-independent, minimize errors, and maximize throughput in VC projects. Our method includes a genetic algorithm where the representation of individuals, the fitness function, and the genetic operators are specifically tailored to get effective policies in a short time. The effectiveness of our method is evaluated with SimBA, a Simulator of BOINC Applications. Contrary to manually-designed scheduling policies that often perform well only for the specific project they were designed for and require months of tuning, our resulting scheduling policies provide better overall throughput across the different VC projects considered in this work and were generated by our method in a time window of one week.