Research
Landing Site Search & Evaluation for Autonomous Medical Evacuation
In this project we 
developed an algorithm for autonomous medical evacuation. We used a full-size helicopter to collect sensor data and test our algorithm at several test sites around Pittsburgh. The algorithm has two stages: First we perform a coarse optimistic evaluation of the terrain. A second stage performs a fine evaluation using a 3D surface of the helicopter. The best landing site is then chosen based on some cost metric.
Link to the homepage of this project.
Collision Avoidance and Planning for Unmanned Aerial Vehicles
The goal of this project was to develop a helicopter that is able to fly between a series of spare waypoints autonomously. The altitude however is so low that there are many obstacles present. Some of the obstacles that we avoided were thin wires, trees, buildings and containers. The helicopter is completely autonomous. It senses the world with a laser rangefinder and a state estimator and sends velocity commands to a controller.
In this paper are more details about our approach.
Link to videos about collision avoidance
Autonomy for Micro Aerial Vehicles
Link to videos about collision avoidance
Link to video demonstrating wire capture
Neural Network Controller for Unmanned Aerial Vehicles
For our machine learning project we developed a online learning controller. It was based on a series of papers at GA-Tech and basically consists of a PD-controller with a neural network as the adaptive element. As the helicopter experience unknown disturbances the controller will adapt and respond. We implemented this project on a Blade CX helicopter.
Link to video of the controller in action
Learning Obstacle Avoidance Parameters
In this project we learned obstacle avoidance parameters for a reactive obstacle avoidance control law from human observation. A subject drove to a prescribed goal while trying to avoid obstacles. We then fit the parameters of a reactive algorithm to best match the training paths.
Paper about learning obstacle avoidance parameters
Link to the Homepage of the Project
Motion and Sensor Simulation for Tartan Racing(Urban Challenge)
The aim of this project is to develop a simulator for the Urban challenge. The simulator will be used to test and develop the algorithm and will aid in visualizing status data. The simulator is structured in a distributed fashion to split processing among several computers. It has the ability to handle an arbitrary number of clients and performs collision and constraint checking.
Link to the Urban Challenge homepage.