Giorgos Mamakoukas | Digital Portfolio

Projects

Research

In the Interactive and Emergent Autonomy Lab , my research focuses on algorithmic development and computational methods for real-time system identification and nonlinear (model-based and data-driven) control of underwater robotics. This work often involves

More specifically, I have developed methods for controllability-based nonlinear control with guarantees for convergence and obstacle avoidance and techniques for linear embedding of nonlinear dynamics with explicit error bounds on model accuracy for the purposes of data-driven control. I enjoy investigating the properties and relations between different optimization and machine learning algorithms. I am currently working on physics-based identification and control of dynamics.

Stable Machine Learning for Prediction and Control

This project focuses on developing machine learning techniques that improve both short- and long-term prediction accuracy. We develop numerical methods that guarantee the stability of data-driven modeling, prediction, and control and can run in real time. I test these methods in simulations and experiments on various robotic systems and applications, including predicting the gait cycle of a biped, stabilizing a quadrotor, and tracking trajectories or pushing blocks using the Franka Emika robot.

Data-Driven Control of Robotic Fish

This project focuses on real-time system identification and data-driven control of robotic fish. I developed a systematic, data-driven methodology that creates linear representations of nonlinear systems that bounds the model accuracy and enables real-time control synthesis. In collaboration with Michigan State University, I tested the approach on tail-actuated robotic fish and show that it ourperforms alternative well-tuned feedback schemes.

Real-Time Control for Nonlinear Systems

This project focuses on developing real-time feedback policies for nonlinear systems. I developed a real-time controller that has a closed-form expression and controllability-based guarantees for convergence and obstacle avoidance.