At the end of the 15 th century, a man by the name of Leonardo da Vinci grew an inspiration to realize one of the oldest engineering fantasies as he attempted the design of a flapping-wing flying machine. Even more amazing is the fact that more than five hundred years after da Vinci's initial work, flapping-wing flight still remains an open problem. Although humans have had the capability of flight for about a century, with the help of various aircraft, we have yet to fully understand or mimic the phenomenon of bird flight.

Modern aerodynamic theory is based on separating the roles of lift and propulsion by attaching engines to rigid aerodynamic frames, as can be seen on commercial aircraft with large thrusters mounted beneath each wing. The engines provide the necessary propulsion that is required to obtain lift from the rigid wings, which allows the use of conventional aerodynamic theory based on fixed wings in a steady airflow. Flapping wings, on the other hand, differ significantly in that they create both the lift and propulsion necessary to sustain flight while interacting with highly unsteady airflows. In this context, conventional aerodynamic theory breaks down due to the complex interactions between unsteady airflows, flexible wing membranes, and high degrees of under actuation.

Given that the dynamics of flapping-wing flight remains an incompletely understood theoretical topic, machine learning becomes an attractive candidate for addressing this type of problem. Using tools from machine learning, we hope to exploit the natural dynamics of flapping-wing flight without requiring an explicit model of them, and moreover, design a robotic bird that continually learns and improves its flight performance through experience. More on the project here.

During the 2003-2004 semesters, I was a member of the Computational Learning and Motor Control Lab (CLMC) at USC, under the supervision of Dr. Stefan Schaal. My research emphasis centered around dynamic motor control of redundant robots. Redundant robots essentially posses more degrees of freedom than are required to execute a given task. This means there are an infinite number of ways that a robotic arm, for example, can generate joint trajectories that will allow it to track a given target with its fingertip. This becomes an extremely important issue when dealing with high degree of freedom systems such as humanoid robots.

I was a student intern for two consecutive summers at the MIT Lincoln Laboratory. My work during the first summer involved working with intrusion detection systems, which are systems that monitor computer network traffic and generate alerts in the event of a suspected attack on the network. I focused on understanding the behavior of these systems for the purpose of modeling their activity and simulating a large network of detection sensors. During my second summer, I implemented a kernel module for the Linux operating system. The purpose of this module was to detect network attacks originating from buffer overflows in which the attacker can modify the return address of the program to point to his supplied malicious assembly code.