Mohamed Naveed Gul Mohamed

Bio

I hail from Tiruchirappalli (Trichy), Tamil Nadu, India. After completing my schooling at Campion Anglo-Indian Higher Secondary school, I pursued my bachelor's in instrumentation and control engineering at the National Institute of Technology (NIT) Trichy.

During my undergraduate studies, I got interested in robotics and joined the robotics club of NIT Trichy - Robotics and Machine Intelligence (RMI). RMI gave me the opportunity to work on interesting projects and learn the tools necessary to build robots. Moreover, I got the opportunity to work with incredibly smart and insanely hardworking people, who also became my best friends for life. We also conducted workshops to teach robotics concepts to interested students, which is where I discovered my love for teaching.

To further my technical knowledge in robotics and artificial intelligence, I did a couple of internships at the Indian Institute of Science (IISc), Bengaluru, and the Indian Institute of Technology (IIT), Madras. I gained valuable insight into how to conduct research and develop solutions in a rigorous way during these internships which helped me decide on where to go forward in my career. It also helped me narrow down to the field of control systems for robotics.

Having discovered my passion for research and teaching and my love for robotics and control systems, I applied to graduate school and got into the Department of Aerospace Engineering at Texas A&M University, College Station. I started working as a graduate research assistant at Estimation, Decision and Planning lab with Dr. Suman Chakravorty, who introduced me to optimal control theory and on which I continue to work to date.

In graduate school, I have mainly focused on optimal control, especially for nonlinear dynamical systems, and extending the theory to data-based/reinforcement learning applications. I initially considered finite horizon fully observed problems. Then, I extended the theory to partially observed output feedback problems. More recently, I have also developed theoretical results for infinite horizon fully observed problems to guarantee global asymptotic stability.

Apart from my main research, I worked on several projects on the side. I worked on building an autonomous pallet jack and developed the entire navigation stack for it during my first couple of years. During my second and third years, I worked on distributed state estimation for unreliable networks. I also got the chance to work on developing reconstruction algorithms for Synthetic Aperture Radars. Towards the end of my PhD, I worked on developing the mathematical reasoning to use dynamic mode decomposition for system modeling, and its relationship to Fourier and Koopman operators. This technique was used in data-driven modeling of periodic orbits in LEO, GEO, and Cislunar space.

Education

Ph.D. in Aerospace Engineering, Texas A&M University, College Station, Texas

Aug '18 - Dec '24

B.Tech. in Instrumentation and Control Engineering, National Institute of Technology Tiruchirappalli (NIT-T), India

July'14 - May '18

Work experience

Estimation, Decision and Planning Laboratory (EDPLab)

Graduate Assistant - Research (Sept '18 - Dec '24)

X , the moonshot factory (formerly Google X)

Control Algorithms Research Intern (June - Aug '21)

Worked on Taara - the moonshot aiming to expand global access to fast, affordable internet with beams of light
Modeled subsystems of an existing free-space optical communication (FSOC) system with modern control methods
Built a platform-agnostic framework to simulate and control the states of the FSOC link
Analyzed the existing control system and proposed an improved version that can give better link performance in certain challenging situations

Reseach work

Optimality and Tractability in Stochastic Optimal Control for Nonlinear Systems

Proved that the deterministic feedback law is near-optimal to the fourth order of a perturbation parameter to the stochastic optimal control problem
Proved that the nonlinear optimal deterministic feedback law can be obtained by solving the deterministic optimal control problem repeatedly in a model predictive control (MPC) type fashion for certain class of systems
Developed a result that showed that the terms in the stochastic feedback law are coupled with the higher order terms, making it hard to compute the stochastic feedback law accurately

Linear System Identification based on Information-State

Developed a novel system identification technique that can model the transient and steady-state response of linear time-invariant and time-varying systems using an autoregressive model and the information state
Compared the proposed technique with other methods and showed that it reduces the need for experimental data by 50\% and is also computationally efficient
Evaluated the technique for identification of nonlinear systems around a trajectory and showed that it outperformed the state-of-the-art methods by two orders of magnitude in the error

Development of an Autonomous Pallet Jack for Warehouse Automation

Developed the entire navigation stack for CAN communication, path planning, estimation and control in ROS
Evaluated path tracking controllers namely, MPC, Pure-Pursuit and PID and achieved < 6 cm tracking error
Assisted in development of a robust localization framework using structural features in the environment, that can achieve better estimates than the state-of-the-art indoor localization frameworks

Data-based Control of Partially Observed Nonlinear Dynamical Systems

Developed a novel method to convert the partially observed problem into a fully observed problem using the information-state and proved that the conversion is optimal after an initial transient
Proved that the nonlinear partially observed problem has a unique solution under some mild assumptions

Efficient Distributed State Estimation over Unreliable Networks

Implemented an algorithm that unifies consensus and covariance intersection methods to produce consistent conservative estimates for robot localization using multiple observers
Evaluated it extensively to show its scalability in terms of state dimension and number of agents and compared it with competing algorithms
Examined the filter’s convergence rate and consistency with changes in graph connectivity