Computer Infrastructure Support for Leveraging Egocentric and Allocentric Representations for Naviga
Abstract
This DURIP proposal is intended to obtain funds to purchase computer equipment to support a current award from the Science of Autono,my section of the ONR. The project title is Leveraging Egocentric and Allocentric Representations for Navigation (LEARN). Current au,tonomous vehicle systems -- service robots, autonomous cars, and unmanned drones -- cannot navigate well in complex, unknown or chan,ging environments. In contrast, mammals navigate flexibly in the 3D natural world. Our project involves functional MRI (fMRI) record,ing of human brain activity while subjects navigate through a large virtual world. These data will be used to create high-dimensiona,l encoding models that reveal how dozens of different navigation-related feature spaces are represented in the brain during naturali,stic navigation. Insights gained from these models will then be used to help develop a new bio-inspired computational framework for,autonomous vehicle navigation. The resulting framework will provide a platform for ground and aerial robots to explore and navigate,in novel, dynamic environments.My component of the MURI award covers the human fMRI experiments on naturalistic navigation and subs,equent modeling of the brain activity data. This modeling component is a prototypical big data problem. Each subjects data consist,of a long time series of brain activity measurements recorded at tens of thousands of different brain locations, along with all of t,he video and behavioral data. These data are modeled in terms of several dozen distinct navigation-related feature spaces that toget,her may consist of thousands of distinct stimulus- and task-related features. Each subjects data is processed independently, so tha,t each one serves as a complete replication of the experiment.Modeling the data in this way provides an enormous amount of informati,on but it also requires substantial computing power. We see funds to purchase additional computer equipment to support this effort.,Tests in our lab show that working with GPUs is 20-50 times faster than working with CPUs. Thus, our main request is to purchase a G,PU compute cluster to support this work.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 05, 2022
- Source ID
- N000142212217
Entities
People
- Jack L. Gallant
Organizations
- Office of Naval Research
- United States Navy
- University of California Regents