Kwangyul Baek

Autorotation of an Unmanned Helicopter by a Reinforcement Learning Algorithm

Autorotation maneuver requires time-critical control inputs and includes highly nonlinear dynamics. Reinforcement learning is a feasible approach that can be applied to this problem. The Q-learning is selected as a reinforcement learning algorithm and a radial basis function (RBF) network is applied as a function approximation technique to handle a large state space in the autorotation problem. Weights of RBF are updated by a back-propagation technique using a direct gradient method. The proposed reinforcement learning algorithm is evaluated by simulations based on a point-mass model of a modified OH-58A helicopter. Nomenclature T = thrust D = drag W = weight of the helicopter V = velocity of the helicopter u = horizontal velocity w = sink rate θ = angle which V makes with the horizon α = angle between thrust vector and vertical axis TPP = rotor tip path plane h = altitude Ω = angular rate of the rotor blades

Mode changing tracker for ground target tracking on aerial images from unmanned aerial vehicles (ICCAS 2013)

This paper addresses mode changing tracker that has global and local tracking mode for efficient target tracking in aerial images from unmanned aerial vehicle. There are two modes in this tracker; Global tracking for object detection and local object tracking. In global tracking, an object in current image sequence is detected with covariance matrix matching. The covariance matrix is one of the efficient ways describing models as fusion of spatial and statistical properties of features. In local tracking, tracker conducts object tracking with kernel-based object tracking algorithm. Kernel-based object tracking algorithm, also known as mean shift, is one of the modern object tracking approaches. We demonstrate the performance of the tracker on aerial image sequences.

Construction of integrated simulator for developing head/eye tracking system

This paper describes the development of integrated head and eye tracker system. Head tracker is performed vision based and it has 7 mm error in 300 mm translation. The epipolar method and point matching are used for determining a position of head and rotational degree. High brightness LEDs are installed on helmet and the installed pattern is very important to match the points of stereo system. Eye tracker also uses LED for constant illumination. A Position of gazed object (3 m distance) is determined by pupil tracking and eye tracker has 1~5 pixel error. Integration of result data of each tracking system is important. RS-232C communication is applied to integrated system and Semaphore method is used for process synchronization.

ADS-B based Trajectory Prediction and Conflict Detection for Air Traffic Management

The Automatic Dependent Surveillance Broadcast (ADS-B) system is a key component of CNS/ATM recommended by the International Civil Aviation Organization (ICAO) as the next generation air traffic control system. ADS-B broadcasts identification, positional data, and operation information of an aircraft to other aircraft, ground vehicles and ground stations in the nearby region. This paper explores the ADS-B based trajectory prediction and the conflict detection algorithm. The multiple-model based trajectory prediction algorithm leads accurate predicted conflict probability at a future forecast time. We propose an efficient and accurate algorithm to calculate conflict probability based on approximation of the conflict zone by a set of blocks. The performance of proposed algorithms is demonstrated by a numerical simulation of two aircraft encounter scenarios.

Vision-based target motion estimation of multiple air vehicles using unscented information filter

This paper presents target motion estimation for multiple air vehicles. The motion of the airborne target is estimated by vision information from camera sensors fixed to the followers. Each camera sensors provides information of the target by three angles: an azimuth angle, an elevation angle, and a subtended angle. In the frame work of information filter, each follower estimates position, velocity and size of the target while communicating own information with other vehicles. The information filter is suitable for decentralized multiple sensor estimation. Due to nonlinear characteristic of vision-based estimation problem, an unscented information filter is applied to target motion estimator. The unscented information filter provides the accuracy and robustness by embedding the unscented transform of the sigma point filter into the frame work of the information filter. The local filter propagates the target state and covariance through a statistical linear error propagation and updates by linear combination of the local information contribution terms. The simulation results demonstrate the performance of the proposed target motion estimator using the unscented information filtering of the three followers.

Decentralized target geolocation for unmanned aerial vehicle with sensor bias estimation

This paper deals with the decentralized approach of target geolocation and sensor bias estimation for multiple unmanned aerial vehicles with bearing angle sensors. The bias of bearing sensor is crucial source that impoverish accuracy of target geolocation. The decentralized estimation approach utilizes the information filtering and dual estimation. The local estimator running in each vehicle estimates the target motion and its sensor bias simultaneously in dual estimation framework. The dual estimation consists of two parallel filters, which are the state filter for target motion and the parameter filter for sensor bias. The information increments of target motion in local vehicles are shared with other vehicles in information filtering framework which is suitable for multiple sensor estimation than conventional Kalman filter. Performance comparison of the proposed decentralized geolocation algorithm with bias estimation with the centralized approaches is demonstrated by numerical simulation.

Developing Head/Eye Tracking System and Sync Verification

This paper describes the development of integrated head and eye tracker system. Vision based head tracker is performed and it has 7mm error in 300mm translation. The epi-polar method and point matching are used for determining a position of head and rotational degree. High brightness LEDs are installed on helmet and the installed pattern is very important to match the points of stereo system. Eye tracker also uses LED for constant illumination. A Position of gazed object(3m distance) is determined by pupil tracking and eye tracker has 1~5 pixel error. Integration of result data of each tracking system is important. RS-232C communication is applied to integrated system and triggering signal is used for synchronization.