The proliferation of adversary undersea capabilities across the Indo-Pacific region has accelerated the need for scalable, intelligent unmanned systems capable of operating collaboratively in contested maritime environments. Large swarms of autonomous underwater vehicles (AUVs), guided by coordinated sensor fusion and adaptive behaviors, represent a critical emerging capability for undersea domain awareness — particularly for the defence of high-value underwater assets across the vast and strategically significant INDOPACOM theatre.
In response to this operational imperative, the Office of the Secretary of Defense has initiated the Resilient and Autonomous Artificial Intelligence Technology (RAAIT) program, designed to integrate collaborative data feeds and AI-driven decision-making into experimentation frameworks. A key proving ground for this initiative is Talisman Sabre 2025, the biennial US-Australia joint exercise, which provides a Live/Virtual/Constructive (LVC) training environment enabling rigorous testing of science and technology capabilities alongside operational military activities. This Australian-hosted exercise offers a uniquely relevant context for validating undersea autonomous systems within a credible Indo-Pacific threat environment.
This paper presents lessons learned from the development and deployment of collaborative AUV swarms within the RAAIT architecture during Talisman Sabre 2025. AUV tactical behaviors were developed using the Advanced Framework for Simulation (AFSIM), coupled with an AI/ML engine to generate, refine, and validate coordinated counter-underwater vehicle (C-UUV) behaviors. These behaviors enabled AUV teams to autonomously detect, classify, and track adversary underwater assets, providing early warning and cueing for the protection of critical undersea infrastructure and high-value platforms.
A central contribution of this research is the demonstrated efficiency of a Modeling and Simulation (M&S)-to-live pipeline, wherein AI-trained behaviors developed in simulation were transitioned and validated on live unmanned platforms with significantly reduced development timelines. The paper examines the fidelity gap between simulated and real-world performance, identifying key variables that influence behavioral degradation or enhancement when transitioning from virtual to live environments. Findings highlight both the transformative potential and current limitations of simulation-trained autonomy in operationally realistic undersea conditions.
The results affirm that integrating M&S environments with AI/ML-driven behavior generation substantially compresses the experimentation cycle, accelerating the delivery of tactically relevant autonomous capabilities to warfighters. For Australia and partner nations operating within INDOPACOM, this approach offers a scalable and cost-effective pathway to building undersea autonomous capacity capable of countering near-peer threats across the region's complex maritime geography.