Distributed goal reasoning for rapid and robust autonomy
Abstract
This proposal aims to contribute to the goal of NRL#s Navy Center for Applied Research in AI (NCARAI), directed by Dr. David Aha, to extend the use of goal reasoning (GR), a framework for autonomous reasoning NCARAI has studied since 2009, from the control of single or few homogeneous autonomous systems to the control of heterogeneous distributed autonomous systems. Goal reasoning is "the process by which intelligent agents continually reason about the goals they are pursuing, which may lead to goal change and has been studied, using different terminology, in multiple disciplines for several decades" (Aha, 2018). GR is useful in complex and dynamic environments, where it is impossible or infeasible to specify upfront which goal should be pursued in each contingency.A distributedrather than centralized architecture should allow collaborating multiagent GR systems to respond more quickly to their local environment, and potentially be more tolerant of communication losses. Our main hypotheses are that distributed GR (a) enables faster reaction to dynamic events than a centralized control strategy and (b) it makes systems more tolerant of communication losses. However, how should a distributed set of GR systems collaborate in dynamic environments?This question needs to be answered by addressing the following research questions (RQ):RQ1. Distributed reasoning: What methods exist for distributed GR systems? What performance measures are used to assess them and how do they compare? Is there anything unique about GR systems in the context of multi-agent (distributed) systems, or should the focus of this question instead be on multi-agent systems, more generally?RQ2. Applications: Do any other applications of GR exist for different domains, such as the space domain, that can help frame the project#s approach?RQ3. GR Stateof the Art (SotA): What is the SotA, especially for continuous ranking of goals, goal discovery, opportunistic reasoning, etc., during mission execution?RQ4. Belief-Desire-Intention (BDI) architectures (NCARAI is using one to continuously rank goals in their GR agent): What BDI methods currently exist that provide guarantees on their performance? Are some of such methods specifically designedfor the characteristics of distributed GR tasks (e.g., partial observability or dynamic events)?To answer these questions and test the hypotheses, we will carry out a literature review and pursue the following innovation objectives (O), using a technical approachbased on software simulations and experimental studies:O1. The realization of a few systems, heterogeneous in terms of sensing, communication, and/or acting capabilities, representing some categories of Unmanned Vehicles of the example scenarios described in the proposal, such as Unmanned Surface Vehicles, low- vs high-sensor Autonomous Underwater Vehicles, or Autonomous Aerial VehiclesO2. The placing of teams of such heterogeneous systems in a simple dynamic environment, possibly basedon a 2D grid world, where they have a common missionO3. The realization of a simple agent architecture, able to execute GR for changing goals according to possible opportunities or discrepanciesO4. The design and implementation of one or more distributed GR algorithms enabling a distributed set of GR systems to collaborate in the dynamic environmentO5. The implementation of one or more existing centralized algorithms to compare with the distributed GR algorithmsO6. The testing of the hypotheses about the distributed GR algorithms through the iterative execution of several experiments, based on varying conditions regarding the dynamic features of the environment (e.g., partial observability, communication disruptions), followed by the statistical evaluation of the results and the revision/improvement and iterative testing of the GR mechanisms.The planned project outcomes are new algorithms, software, experimental data, and scientific publications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 15, 2024
- Source ID
- N629092412056
Entities
People
- Paola Rizzo
Organizations
- Office of Naval Research
- United States Navy