Research Project Title:
Online HD Map Generation For Autonomous Vehicles
abstract:Automated vehicle systems often rely on high definition (HD) maps that encode information about the static environment surrounding the vehicle, such as the relative location and distance metrics of lanes, curbs, street signs, and other objects. These HD maps are transformed into the vehicle reference frame at run time to aid in the downstream planning and decision processes of the automated vehicle. My research centers on leveraging state of the art object detection networks and graph neural networks (GNNs) to generate these HD maps on the fly. Our deep learning model will take in data from prior SD maps, lidar pointclouds, and camera images to infer both a detailed static road map and a temporal object relation graph to account for dynamic changes within the scene. Our overarching goal is to create a more accurate and detailed representation of the vehicle’s environment than any other existing HD map generation model. This will allow autonomous vehicles to more effectively navigate their surroundings and make better decisions, ultimately leading to safer and more efficient operation.
About:
Hi! I'm a fourth year computer science student at MIT who spent the past two years taking a deep dive into ML research with Professor Daniela Rus and Alexander Amini. As a past startup founder, my interests lie more within industry and the entrepreneurship sphere rather than just pure academia. I aim to leverage my research to either advance an existing product in the industry or innovate and build out a novel product for a startup. My research this year especially excites me because I can envision the technology that we developed to eventually scale and be used by autonomous vehicles all across the globe.
