Real-Time, Low-Power Edge Intelligence on Mobile Robotic Platforms using a Neuromorphic Computing Architecture

Abstract

Deep networks have demonstrated immense promise for enabling autonomy. However, the current paradigm of using stream-computing architectures for inferring network responses is ill suited for many types of robotic platforms, especially those performing long-term missions with limited battery lifetimes. This is because stream-computing algorithms are often incredibly power intensive. Going forward, it is crucial to retain the universality and flexibility of non-linear-inference capability of deep networks, while reducing their overall power requirements. Doing so will help design a new generation of intelligence system for the internet of things, also called edge computing. For the Navy this would permit on-platform processing of sensor feeds and decision making for navigation. In this grant, we will investigate the use of neuromorphic inspired computing to enable real-time, low-power edge computing. We will leverage and extend an existing neuromorphic hardware architecture for processing spatio-temporal sensor feeds and subsequently transforming those analyses into autonomous behaviors. We will show that this distributedarchitecture can effectively address a variety of tasks: (1) It innately allows for continuous low-shot learning of new object classes without catastrophic forgetting; (2) It can facilitate transfer learning between domains for problems like semantic segmentation, change detection, and linguistic scene summarization; (3) it will allow for close loop vision based navigation. We will illustrate that the proposed brain-inspired models can be implemented to perform unsupervised learning, which will be helpful for handling massive amounts of non-curated data. We will additionally integrate this neuromorphic hardware architecture with a robotic platform to illustrate the practical utility of edge computing.We will demonstrate that on-platform object detection and recognition is possible, with high accuracy and low uncertainty, in challenging environments.We will also emphasize advanced autonomy behaviors. All of these capabilities will be a boon for underwater survey platforms performing tasks like long-termsurveillance, swarming and counter-swarming with vehicle persecution, and mine detectionand neutralization.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2023
Source ID
N000142312084

Entities

People

  • José Príncipe

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Florida

Tags

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.
  • Unmanned Aerial System (UAS) Autonomous Capabilities and Mission Reconnaissance.

Technology Areas

  • 5G
  • 5G - Internet of Things
  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks
  • Autonomy
  • Autonomy - Autonomous System Control