Projects

---

Our first project was an entry in the 2017 NASA Aeronautics University Design Challenge, wherein we created a white paper concept of Goldeneye AB1, a supersonic business jet designed to meet specific performance and environmental criteria for functionality and efficiency. Our concept included a novel variable-geometry wing that was designed to change shape to optimize lift and reduce drag in both supersonic (cruise) and subsonic (takeoff and landing) flight regimes. My main contributions to the paper were the design of the fuselage and the mechanism that controlled the variable wing geometry, in addition to being the principal editor of the paper. 

A group of entirely freshmen and sophomores, we were thrilled to be awarded an honorable mention, especially considering that we competed against much older, more experienced teams from schools with dedicated aerospace departments (many of whose submissions were capstone senior projects). NASA flew us out to Langley Research Center to accept our award, present our work, and tour the storied facilities. Our paper is attached below.

Our next project was fully funded by Aptiv PLC, an automotive parts manufacturer involved in autonomous vehicle technology. We also collaborated with Prof. Francesco Borrelli's Model Predictive Control (MPC) Lab, which has built the Berkeley Autonomous Race Car (BARC), a 1:10 scale RC car designed to navigate autonomously while performing many complex maneuvers (drifting, lane changing, object avoidance, etc.). We took this platform and updated the sensor suite to include high resolution cameras and a LIDAR sensor. 

Lane Keeping

We employed a state feedback controller for the vehicle's lane keeping system. The image feed from the vehicle's camera is sent to a separate computer, where an algorithm detects and classifies the lane. This information is sent back to the vehicle which adjusts its motion to keep its distance from the center of the lane small. 

Object Detection

Our object detection system relied on sensor fusion, combining the information from the camera—which is good for object detection/classification—with the LIDAR's data, which is used to measure distances from other objects. This approach is necessary because it is insufficient to simply detect where an object is in a two dimensional field of view, and information about depth is needed as well.

Motion Planning and Obstacle Avoidance

We designed several algorithms to control the vehicle's velocity, steering angle, and lateral position. Our controllers used Model Predictive Control algorithms to plan and execute trajectories both in the presence of and without obstacles. 

Aptiv invited us to present our work, demo our car's abilities, and tour their facilities in Mountain View. Our presentation is included below.