RJ321-G1

Despite the enormous potential of battery-powered drones in the Air Force and commercial applications, true autonomy is still not achieved. A pilot constantly navigates the drone using on-board cameras and real-time wireless communication. All components constantly drain battery, and flight time is currently well below an hour. Constant communication makes the drone vulnerable to adversaries. To extend flight time and improve drone stealth capabilities, there is a need for computer systems that can make drones autonomous. Drone operation and reaction require the onboard computer system to accomplish perception and control. These tasks rely heavily on machine learning and stochastic control algorithms. These algorithms are sig nificantly compute intensive and cannot be accomplished with current general-purpose platforms within the power and payload constraints of battery-powered drones. Without innovative compute systems, such capa bilities and true autonomy may remain out of reach. This limitation is due to the diminishing benefits from traditional transistor scaling and stagnant improvements in general-purpose computing. To tackle these chal lenges, this project aims to develop a holistic and cross-stack solution-from programming languages down to circuits-that enables perception and control in energy-constrained drones. The key innovation is a many- core accelerator that efficiently executes machine learning and stochastic control algorithms. To support a wide variety of algorithms, we leverage the insight that many of these algorithms are stochastic optimiza tions that can be effectively expressed using mathematical formulations. Using this insight, we embark on a basic scientific investigation that develops a domain-specific language for learning and stochastic control; an interpreter that maps the specified algorithms to the instructions of a novel virtual machine; a compiler that schedules and translates the virtual machine instructions to the accelerator reconfiguration; a design of a novel clustered many-core accelerator; operating system primitives that integrate the computing platform; FPGA prototypes; and finally deployment on quadcopters.

San Diego University Of California

$217,262.00

Feb. 20, 2019

La Jolla, California 92093-0934 United States

Georgia Tech Research

FA95501710274

Tas::57 3600::Tas '(Yip) Accelerated System Design for Perception and Control in Energy-Constrained Uavs'

12.800 - Air Force Defense Research Sciences Program

Air Force Office of Scientific Research (AFOSR) [DoD - USAF - AFMC - AFRL]

Project Grant

Atlanta, Georgia 303185775 United States