Neuromorphic Architectures for Fast Low-Power Robot Perception, Work Unit IT015-09-41-1G25
Technical Report,01 Oct 2016,30 Sep 2019
NAVAL RESEARCH LAB WASHINGTON DC WASHINGTON United States
Pagination or Media Count:
This memorandum summarizes the efforts of work unit 1G25 where new brain-inspired neuromorphic computing technology and deepconvolutional neural networks CNN where applied to develop real-time low SWaP scene understanding capabilities for mobile robotic systems. Specifically, we sought understanding of the relationships between the perception task, CNN-based algorithms, and the constraints of neuromorphic systems and to derive principles of CNN design for neuromorphic architectures.