Bernadette Bucher
PhD Student, University of Pennsylania

Title: Uncertainty-based Mapping and Navigation

Abstract

Robots must be able to navigate in previously unseen settings in order
to ultimately operate autonomously in environments designed for
humans.  Challenges in novel environments include navigating to
semantic targets such as chairs and tables outside of the agent'sfield
of view (object goal navigation) and navigating to target coordinates
relative to the current agent location around unobserved obstacles
(point goal navigation). Since the agent does not have access to
occupancy or semantic maps of the environment, this occupancy and
semantic information needs to be inferred to perform these navigation
tasks. Most current methods implicitly infer these unknown maps
through end-to-end policies directly predicting agent actions from
agent observations. In this presentation, I will present our novel
framework for navigating in unseen environments by explicitly
predicting semantic and occupancy maps and leveraging the uncertainty
over our map predictions to make navigation decisions. I will show
experimental results in the visually realistic environments of the
Matterport3D dataset in the Habitat simulator where we demonstrate
improved results on both the object goal and point goal navigation
tasks.

Bio

Bernadette Bucher is a PhD Student in the GRASP lab at University of
Pennsylvania co-advised by Dr. Kostas Daniilidis and Dr. Nikolai
Matni. During her PhD studies, she completed a M.S.E. in Robotics from
University of Pennsylvania in 2020 and was awarded the Haidas and
Chryssikou Fellowship for her research. Bernadetteā' research
interests lie in the intersection of robotics, computer vision, and
machine learning. The focus of her PhD research is on improving the
generalization ability of vision-based models for robotic manipulation
and navigation to novel data. During her PhD, Bernadette interned with
Dr. Dieter Fox in the Seattle Robotics Lab at NVIDIA Research. Prior
to starting her PhD, she was a Senior Software Engineer at Lockheed
Martin Corporation where she worked from 2014 to 2019. Bernadette
received an M.A. in Mathematics, M.A. in Economics, and B.S. in
Mathematics and Economics from The University of Alabama in 2014.