I obtained my Ph.D. in the field of Control Systems under Prof. Roberto Horowitz at UC Berkeley. Here are some highlights of my work as a graduate student researcher:
1) ROBUST FEEDBACK-FEEDFORWARD CONTROL DESIGN – FREQUENCY RESPONSE DATA DRIVEN APPROACH
1) Developed a Robust Control Design Tool: Created an advanced tool for simultaneous design of stabilizing feedback controller and vibration suppressing feedforward controller, leveraging frequency response measurements of the actuators and their vibration interactions
2) Innovative Methodology: Designed a novel a H_2-H_∞ control algorithm tailored for multi-input multi-output systems
3) Data Driven Robustness: Utilized multiple sets of frequency response actuator data and vibration data to create a common feedback controller meeting performance and reliability targets; Convex H_∞ constraints ensure closed-loop stability with good gain and phase margins
4) Vibration Suppression: Designed and implemented a fixed feedforward controller using the data driven approach; Added an adaptive control element to modify the fixed control action; Finally added an input shaper to reduce vibration at source
TRAJECTORY FOLLOWING - MODEL PREDICTIVE CONTROL (MPC) + REINFORCEMENT LEARNING
1) Developed a MPC Simulation Tool: Created an output feedback model predictive control design to follow a pre-defined trajectory with constraints on stroke usage of all control actuators. System Example: Highway lane changing
2) Reinforcement Learning Approach: Trained a droid model based on the same pre-defined reference trajectory using a four layered neural network; Updated the training set with measurement data from the MPC tool
3) Implementation: Developed an explicit closed form MPC solution to implement this method on a hard disk drive (HDDs); HDDs have limited computation power and it is imperative to solve majority of the control problem offline