Research Project Title:
Non-Line-of-Sight Obstacle Detection for Autonomous Vehicles
abstract:The following work is centered around developing the ability for autonomous vehicles to detect occluded obstacles. Obstacles that are occluded and out of the vehicle’s line of sight can be a major source of collisions for autonomous vehicles or indoor robots. In my proposed project, I will be working on detecting such obstacles and anticipating such collisions by observing the shadows and other changing illumination effects caused by light due to the obstacles’ motion. A sequence of images of a corner that the vehicle is approaching will be fed into a 3-dimensional convolutional neural network to identify the shadows and deduce the manner in which they move. The manner in which these shadows change will be analyzed to conduct higher level reasoning in regards to the motion of occluded objects. This work will first be evaluated on a variety simulated corners and occluded objects to account for different surface and lighting conditions. Once a satisfactory classification accuracy is achieved, the algorithm will be integrated into an autonomous wheelchair to conduct real-world experiments.
I have been interested in robotics and the autonomous vehicles space, for over six years. This project will allow me to develop the domain knowledge necessary to to excel in this field. I feel that my past internships at Uber, BMW, Microsoft Research, and the Carnegie Mellon Biorobotics Laboratory, as well as my enrollment in 6.141 and 6.036 have given me the necessary skills in autonomy and control to succeed with my upcoming research project.