Research Summary for Girish Chowdhary

All of the material on this web page, including the attached publications, have been copyrighted by Girish Chowdhary, co-author/s and the proprietary institute.

Please email me at Girish.Chowdhary(att)gatech.edu me if you would like more information.

Member of Aerospace controls group at Georgia Tech

"Theory and Flight Test Validation of Long Term Learning Adaptive Flight Controller" Chowdhary Girish, Johnson Eric, Proceedings of the AIAA Guidance Navgiation and Control Conference, Aug 2008 USA : In this paper we extend the method and theory of background learning adaptive controller to Single Hidden Layer Neural Networks. Furthermore, we present flight test results that affirm the practical stability of the long term learning control law that utilizes both current and past data concurrently for NN training.

"Adaptive Neural Network Flight Control Using both Current and Recorded Data" Chowdhary Girish, Johnson Eric, Proceedings of the AIAA Guidance Navgiation and Control Conference, Aug 2007 USA : Most current adaptive methodologies which use error based recursive training methods rely only on the instantaneous states in order to tune the adaptive gains. For high dimensional problems commonly encountered in control of high performance systems these methods are susceptible to adapting only on the current region of state space. Since these methods do not exhibit long term learning and global adaptation, little performance gain can be expected when a system returns to a previously encountered region of the state space. In order to incorporate long term learning in the adaptive control architecture, we propose a novel approach to adaptive control which uses the current or the online information as well as stored or background information for adaptation. We show that using a combined online and background learning approach it is possible to guarantee long term learning in the adaptive flight controller, which enhances performance of the controller when it encounters a maneuver that has been performed in the past. We use Lyapunov based methods for showing boundedness of all signals for a presented method. The performance of the proposed method is evaluated in the high fidelity simulation environment for the GTMAX UAS maintained by the Georgia Tech UAV lab. The simulation results show that the proposed method exhibits long term learning and faster adaptation leading to better performance of the UAS flight controller.

"Aerodynamic Parameter Estimation from Flight Data Applying Extended and Unscented Kalman Filter" Chowdhary G., Jategaonkar R., AIAA AFM Aug 2006 : The combined state filtering and parameter estimation problem in recursive parameter identification from real flight data is nonlinear and often handled by the Extended Kalman Filter (EKF) which is accurate only to the first order. To overcome the problems posed by the first order approximation inherent in the EKF the Unscented Kalman Filter (UKF) has been proposed, which propagates carefully selected sigma points through the nonlinearities of the system dynamics. The UKF, which is accurate at least to the second order has been tested extensively in aerospace navigation filters, however its use in aerodynamic parameter estimation from real flight data is relatively unexplored. This paper analyzes the feasibility and possible advantages of using the UKF for recursive parameter estimation by comparing its performance with the EKF and offline estimation methods.

"Control of a VTOL UAV via Online Parameter Estimation", Chowdhary G., Lorenz S., AIAA GNC 2005: Adaptive control via online parameter estimation is achieved using an Extended Kalman Filter, the estimated parameters of the system are used to update a reference model in real-time which is used with optimal linear control methods to achieve stable and robust control in presence of parameter uncertainty, noisy data, and biased measurements.

"Non-Linear Model Identification for a Miniature Rotorcraft, Preliminary Results", Chowdhary G., Lorenz S., AHS 2005: Time domain system identification methods are employed to identify an extended linear model of a VTOL UAV in hover domain using only noisy sensor measurements and recorded pilot inputs. The feasibility of employing time domain methods for system identification for miniature UAV projects is demonstrated. The identified model is demonstrated by testing against real flight data not used in the modeling process.

Undergraduate Thesis: "Control of Smart Space Structures using Estimators", Chowdhary G., RMIT University Australia 2003: A methodology is proposed for Control of Smart space structures operating in a stochastic environment using optimal control methods (LQG) and Finite Element Analysis. Finite Element methods are used for dynamic modeling of complex space structures. Reduced order observers are implemented to account for unobserved states and Kalman filters are implemented to handle stochastic measurements. The methodology is demonstrated by treating two possible classes of Smart structures via theoretical modeling and simulation.

Research Report: "Online Optimal Control", Chowdhary G, DLR Internal Report Nov 2005: The Extended Regulator Problem is defined as updating an optimal regulator for a discrete time-variant linear system as the system parameters undergo change. Optimal regulator is synthesized by solving the Discrete Time Riccati Equation (DARE) at each time step. Alternative methods to the Ordered Schur form based solution to the Riccati equation are analyzed. Computationally more efficient, recursive, online optimal gain updating method are demonstrated and compared with traditional methods through simulation.

Undergraduate Team Research Project: "Conceptual Design of a Manned Mission to Mars", RMIT University, Australia 2002. A systems based conceptual design of a complete manned mission to mars is presented using a two stage mission. My major contribution in the project was the conceptual design of a safety oriented Environment Control and Life Support System (ECLS). I assured safety via implementing a three level redundancy hierarchy, I received the Orsen-Wells award for achievement and innovation in Engineering for this project.