Wednesday, June 17, 2015

2015-06-16: Autonomy Incubator Upgrades Tree-Dodging Demo

As the entire AI team prepares for NASA Deputy Administrator Dr. Dava Newman's inaugural visit to NASA LaRC this Thursday, Dr. Loc Tran has built a veritable artificial forest in which to present his autonomous obstacle avoidance—colloquially "tree-dodging"— machine learning research. So far, nine Christmas trees, three potted palms, one fern relocated from the lobby and one ficus have sprouted up in the indoor flying range.

While obstacle avoidance has always been an area of interest for the Autonomy Incubator, as our Dances With Drones demonstration shows, Loc's work is especially amazing because of just how much autonomy his UAVs command. In addition to path planning, his program also uses information from a front-facing camera on the UAV learns to detect features as it flies, and then it avoids obstacles based on those features. If the UAV makes a mistake, Loc goes over the video and corrects the vehicle's responses. These corrections are used to create a new model, the UAV flies again, and the process gets repeated until the vehicle can safely maneuver through the course. Eventually, with enough training, the UAV will become so skilled that it can navigate a course it's never seen before. Literally, Loc teaches the UAV how to avoid obstacles, and it learns from his guidance.   

During Thursday's exhibition, the UAV will take off with a package (most likely a 3D-printed banana - for scale, of course), autonomously navigate the forest, transition into a "neighborhood", and then... 

What, did you think we were just going to put all the juicy details of our coolest demonstration ever online? Absolutely not. Check back tomorrow for more pictures and more information about the AI's plan's for Deputy Administrator Dava Newman's visit.

No comments:

Post a Comment