Baro Hyun

Deterministic Relative Attitude Determination of Three-Vehicle Formations

DOI: 10.2514/1.42849 This paper proves that deterministic relative attitude determination is possible for a formation of three vehicles. The results provide an assessment of the accuracy of the deterministic attitude solutions, given statistical properties oftheassumednoisymeasurements.Eachvehicleisassumedtobeequippedwithsensorstoprovideline-of-sight,and possibly range, measurements between them. Three vehicles are chosen because this is the minimum number required to determine all attitudes given minimal measurement information. Three cases are studied. The first determines the absolute (inertial) attitude of a vehicle knowing the absolute positions of the other two. The second assumes parallel beams between each vehicle to determine relative attitudes, and the third assumes nonparallel beams for relative attitude determination, which requires range information to find deterministic solutions. Covariance analyses are provided to gain insight on the stochastic properties of the attitude errors and the observability for all three cases.

Persistent visitation under revisit constraints

This work is motivated by persistent Intelligence, Surveillance, and Reconnaissance (ISR) missions, where an Unmanned Aerial Vehicle (UAV) is to perpetually fly over a number of unidentified objects within a given search area and classify the objects based on information collected. The aspects of the problem pertaining to the collection of information and classification are abstracted into a revisit rate for each object of interest which, if respected, yields proper classification. The problem of finding paths that meet these revisit rate requirements is treated and periodicity properties of solutions to the path planning problem are derived. Furthermore, heuristics to solve this problem are presented and, through numerical simulations, the effectiveness of the method is illustrated.

Mixed-initiative nested classification by optimal thresholding

The purpose of this paper is to demonstrate that having two classifiers, a trichotomous classifier (true, false, or unknown) with workload-independent performance that turns over the data classified as unknown to a binary classifier (true or false) with workload-dependent performance, gives superior classification performance (lower probability of misclassification) compared to a single dichotomous classifier. We relate the classifiers performance to the inherent difficulty of the classification task at hand (classifiability), and compare the performance of different classifiers.

Automated classification system for Bone Age X-ray images

Bone Age (BA) determination using radiological images of left hands and wrists is important in pediatric endocrinology to correctly assess growth and pubertal maturation. In this paper, we propose a fully automated Greulich and Pyle Atlas (GP) bone age determination system using feature extraction and machine learning classifiers. The original contributions of this paper are as follows: (i) We use commercially available morphing tools to create a modified GP atlas that has images regularly spaced at three month intervals, (ii) We propose a novel Singular Value Decomposition (SVD) based feature extractor to create a feature vector. We use the Scale Invariant Feature Transform (SIFT) to extract features from the images then apply SVD to compose the feature vectors. Then, we train a Neural Network classifier using the generated feature vectors. Our preliminary results show that, even with a small number of training data sets, we obtain promising results. Future direction is discussed.

Robotic Exploration with Non-Isotropic Sensors

In this paper, we present an experimental implementation of an information-based path planning algorithm utilizing a three-wheeled ground robot. The robot is equipped with multiple ultrasonic sensors with which it collects information about objects of interest located within the exploration area. The mission of the robot is to identify the radius of the objects. This problem is challenging as the sensors are non-isotropic, i.e., the sensors have a limited angle of view, and the sensors are range-based, i.e., the amount of information collected by the sensors depends upon the distance from the sensor to the object of interest. The implementation shown relies on interpreting the information gathered as the inverse of uncertainty. Once a specified level of uncertainty reduction is achieved, the object is considered identified. Experimental results for a couple of different objects are provided. The results show close agreement with predicted simulations.

Optimal Classification by Mixed-Initiative Nested Thresholding

We propose a novel architecture for a team of machine and human classifiers (i.e., a mixed-initiative team). We adopt a model of performance that is workload-dependent for the human and workload-independent for the machine. The team is structured in a nested architecture that exploits a primary trichotomous classifier (returning true, false, or unknown) with workload-independent performance that turns over the data classified as unknown to a secondary dichotomous classifier (returning true or false) with workload-dependent performance. The novel classifier architecture outperforms other classifiers, such as a single dichotomous classifier or a simple nested two-classifier team.

Path planning for optimal classification

As stated in the Office of the Secretary of Defenses Unmanned Aircraft Systems Roadmap 2005-2030, reconnaissance is the number one priority mission for Unmanned Air Vehicles (UAVs) of all sizes. During reconnaissance missions, classification of objects of interest (e.g, as friend or foe) is key to mission performance. Classification is based on information collection, and it has generally been assumed that the more information collected, the better the classification decision. Although this is a correct general trend, a recent study has shown it does not hold in all cases. This paper focuses on presenting methods to plan paths for unmanned vehicles that optimize classification decisions (as opposed to the amount of information collected). We consider an unmanned vehicle (agent) classifying an object of interest in a given area. The agent plans its path to collect the information most relevant to optimizing its classification performance, based on the maximum likelihood ratio. In addition, a classification performance measure for multiple measurements is analytically derived.

Optimally-informative path planning for dynamic Bayesian classification

An agent, consisting of an unmanned aerial vehicle (UAV) carrying strapped-down sensors, is to examine a number of unidentified objects within a given search area, collect information, and utilize that information to classify the objects. The problem is challenging because the mission time is often limited, the agent is only provided with partial a priori information, and the amount of information that the sensor can measure is dependent on the relative position of the agent with respect to the object. Our technical approach is three-fold. First, we model the motion of the agent using a kinematic model with constant altitude. Second, we use a performance prediction model that gives the probability of target discrimination as a function of the range from the sensor to the object. Third, a linear classifier that utilizes Bayes’ theorem diagnoses the status of the objects of interest while an information-theoretic measure is used to quantify the uncertainty in classification. We pose an optimal control problem that minimizes the classification uncertainty while taking differential constraints and the time history of the agent’s steering decisions as the control input. We investigate whether maximizing information by choosing informative paths always minimizes the classification uncertainty.

Sequential bayesian classification decisions for mobile sensors

This work is motivated by the U.S. Air Forces Intelligence, Surveillance and Reconnaissance (ISR) mission, where an Unmanned Aerial Vehicle (UAV), or an agent, is to fly over a number of unidentified objects within a given search area, collect information using onboard sensors, and classify the objects. The problem is challenging because the mission time is limited, the agent is only provided with partial a priori information, and the amount of information that the sensor can measure is dependent on the range and the azimuth of the explorer with respect to the object. A sequential decision problem (path planning) is posed that incorporates the potential loss of the classification outcome that is made by an autonomous moving agent. The problem is solved using stochastic dynamic programming. The resulting path exploits the interaction between the agent kinematics, informatics, and classification. Numerical simulation results that validate the concept are presented.

A revisit-based mixed-initiative nested classification scheme for Unmanned Aerial Vehicles

Unmanned Aerial Vehicles (UAVs), used often by the Armed Forces for Surveillance and Reconnaissance (S&R) missions, are powerful classification agents to inspect objects of interest (OOIs) under human supervision. To achieve improved decision-making, we have previously explored the idea of a two-tiered classification structure, where a primary trichotomous classifier (machine) precedes a secondary dichotomous classifier (human). The trend for future operations is for a single operator to control an increasing number of UAVs. However, low human-to-UAV ratio can result in a stressful situation for the human operator, which is undesirable for successful classification and UAV management. To address this issue, we extend our previous work to a three-tiered classification scheme, where an intermediate revisit sensor makes a decision to revisit the OOI in cases where the primary classifier is unsure, which can be caused by noisy sensor data or viewing from a poor vantage point. We compare the performance (i.e, the probability of misclassification) under single, two-tiered, and three-tiered classifier schemes and show the efficacy of the proposed technique.