RAIDER: Resilient Actionable Intelligence for Distributed Environment understanding and Reasoning
Abstract
This project tackles the need to develop scalable computational methods for joint perception and planning that enable a team of agents to extract task???relevant information from massive, multi???modal data sources to support collaborative and distributed optimal planning with formal performanceguarantees. Future control and decision???making systems need to strike a balance between guaranteed performance in the presence of uncertainty, extraction of information relevant to the task at hand from streaming data, and reduced algorithmic complexity to enable real???time loop closure, re???planning, and learning. This confluence of control, estimation, machine learning, and computational science and engineering is necessary for high???confidence, high???reliability, minimal???supervision autonomous systems that can understand and act in high optempo missions. The proposed technical approach addresses head on the need to develop novel methods for extracting structure and understanding context from massive heterogeneous data in real time, ensuring safe operation in unstructured, dynamically changing conditions, and enabling extemporaneous multi???robot collaboration through compositional and hierarchical organization of knowledge. The synergistic research program is organized along five thrusts. Thrust 1 develops multi???modal representations forunified perception and action that can be used across different modalities and spatio???temporal scales; Thrust 2 develops methods for goal???oriented, task???aware perception, inference, and communication that account for the underlying control objective and the available informational and computationalresources across the team; Thrust 3 focuses on enabling resilience by quantifying uncertainty in the unified multi???modal representations, allowing for principled decisions on information exchange between team members, and developing methods to deal with the presence of adversarial agents; Thrust 4develops hierarchical architectures and compositional models that help solve planning, estimation, and coordination problems at the right level of granularity and, finally, Thrust 5 develops online distributed decision???making and inference methods capable of dealing with streaming multi???resolution data. The proposed research has the potential to redesign the landscape of intelligence, surveillance and reconnaissance, acquisition and processing of military intelligence in future Naval missions. A successful outcome of this research will provide unprecedented modeling and inference capabilities to estimatequantities of interest for the Navy (such as the presence and location of potential threats, target tracking and intent recognition, perception and world modeling capabilities for autonomous system navigation and coordination) using heterogeneous streaming data from sensors (e.g., chemical, visual,inertial, wearables) and from cooperative and adversarial actors (e.g., frequency of communication, twitter feeds, contextual information). These capabilities are crucial for future Naval operations, where high???level situational awareness from massive streaming data needs to be obtained in real time and in adecentralized fashion. Moreover, the proposed research will provide fundamental theory and algorithms for distributed and hierarchical planning that will enable fast and distributed decision???making at mission optempo, capable of leveraging a hierarchy of multi???scale/multi???resolution world models toreduce the computational workload and enable operation on resource???constrained distributed embedded platforms.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 19, 2018
- Source ID
- N000141812828
Entities
People
- Jorge Cortés
Organizations
- Office of Naval Research
- United States Navy
- University of California, San Diego