Autonomous Resilience Agent

A Collaborative Reasoning Framework for Anomaly Diagnostics in Underwater Robotics

Paper (arXiv) Code (GitHub)

Authors

Markus Buchholz, Ignacio Carlucho, and Yvan R. Petillot

School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK

AURA (Autonomous Resilience Agent) is an AI framework for managing underwater vehicles. Through Human-in-the-Loop Distillation, it translates sensor data into problems, collaborates with operators on solutions, and retains expert insights as lasting knowledge-evolving into an adaptive, trustworthy partner for resilient human-robot teams.

AURA Concept

The AURA (Autonomous Resilience Agent) framework functions as a distributed cognitive ecosystem for Anomaly Diagnostics. As illustrated in the diagram below, AURA's core process is to analyze RAW data from the Physical AUV (Observed State), compare it against the Digital Twin (Expected State) to find problems, and turn that complex data into a clear, human-understandable diagnostic output via a collaborative dialog with the operator.

AURA Concept

Architectural Overview Video

This video provides a narrated walkthrough of the AURA: Autonomous Resilience Agent framework, which is a Collaborative Reasoning Framework for Anomaly Diagnostics in Underwater Robotics. It explains how AURA translates complex raw data into clear, insightful problem descriptions for the human operator. The architecture is detailed, showing how the system compares the Physical AUV (Observed State) to its Digital Twin (Expected State) to spot anomalies. Crucially, the video explains the collaborative roles of the two AI agents (Agent A and Agent B) and the Vector Database (VDB) in an innovative learning cycle that significantly improves the diagnostic quality and reduces operator workload after just a single human-AI interaction.

Citation

If you use AURA in your research, please cite our paper:

@misc{buchholz2025aura,
      title         = {A Collaborative Reasoning Framework for Anomaly Diagnostics in Underwater Robotics}, 
      author        = {Markus Buchholz and Ignacio Carlucho and Yvan R. Petillot},
      year          = {2025},
      eprint        = {2511.03075},
      archivePrefix = {arXiv},
      primaryClass  = {cs.RO},
      note          = {Under review at the ICRA 2026.},
      code          = {https://github.com/markusbuchholz/urosa_underwater_autonomy},
      url           = {https://arxiv.org/abs/2511.03075}
}