YIP Graph Learning for Distributed Situational Awareness with Limited Communication and Localization

Abstract

Future naval operations will increasingly rely on the collaboration of multiple autonomous robots. However, contested environments limit the communication and localization abilities of robot teams, creating a critical obstacle to multi-robot coordination. To address this pressing concern, we propose leveraging the power of Graph Neural Networks (GNNs) to enable asynchronous coordination in without global frames of reference. By harnessing the inherent structure of multi-robot teams via graph representations and integrating insights from the multi-agent control literature, we will leverage GNNs to enhance multi-robot coordination capabilities. This proposal outlines a comprehensive framework that harnesses the potential of GNNs to enable robust, adaptive, and resilient coordinationamong unmanned systems, ensuring the Navy s continued effectiveness in complex and challenging scenarios.Our technical approach leverages the strengths of GNNs to improve multi-robot coordination, and the strengths of multi-robot coordination to improve GNN performance. Shared situational awareness among multi-robot teams requires frequent communication. Existing approaches cannot reliably handle lost communication over long horizons. We will cast the synchronization problem as a problem of predicting a path on a graph, allow us to find a solution with GNNs. On the other hand, multi-robot coordination rules are designed to coordinate without a local reference frame, something that GNNs are not designed for. In this case, we will enforce reference frame-invariance on GNNs, guaranteeing that they are able to be employed for coordination in robots own local reference frames without require external positioning. Results will be demonstrated for distributed search and tracking of a moving target in a marine environment in simulation as well ason robotics platforms in an outdoor body of water.Combining these advances will enable teams of robots to coordinate in contested environments when communication is difficult and external positioning is unavailable. Such requirements are especially important for the deployment of autonomous Uncrewed Underwater Vehicles, but apply equally well to the deployment of large aerial swarms and otherautonomous teams of robots.

Document Details

Document Type
DoD Grant Award
Publication Date
Apr 10, 2025
Source ID
N000142512225

Entities

People

  • Kevin Leahy

Organizations

  • Office of Naval Research
  • United States Navy
  • Worcester Polytechnic Institute

Tags

Fields of Study

  • Computer science

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Neural Network Machine Learning.
  • Robotics and Automation.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - DoD AI Strategy
  • Autonomy
  • Autonomy - Autonomous System Control
  • Autonomy - UAVs