Reinforcement learning for autonomous marine platforms.

Research § soundings

• 2025 — 2026 MSc thesis · UPC × ICM-CSIC

  Multi-agent RL for marine hazard mapping

  Cooperative boundary tracking of harmful algal blooms and oil spills with decentralized agents under local sensing and limited communication. TransfQMix policies evaluated across observation modes, communication loss, sensor noise and team-size transfer.

  Project page & experiments explorer
• 2025 — present Vision + RL · ICM-CSIC × UPC

  Camera-only localization of small floating buoys

  A hybrid sim-to-real pipeline: synthetic + real data feeding a YOLO detector, paired with a PPO policy with curiosity and LSTM memory, trained in a purpose-built high-fidelity simulator. Trained RL policies with YOLO detection have already been validated on the real BlueBoat.

  Project page & demo
• 2026 — present Robotics engineering · ICM-CSIC

  Custom BlueBoat ASV platform

  Development of a Blue Robotics BlueBoat as a fully autonomous ocean-monitoring platform — leading the autonomy, onboard perception and 3D mechanical design, and contributing across electronics, ROS2, Ardupilot and on-edge AI inference. A SITL simulation of the platform lets policies be tested safely before water trials, and RL control is already running on the real vehicle.

  

  ROS2 Ardupilot SITL Edge inference 3D design

Publications § log

• IEEE ICRA

  Multi-Agent Reinforcement Learning for Autonomous Mapping of Marine Environmental Hazards

  Decentralized TransfQMix policies for cooperative boundary tracking of algal blooms and oil spills; evaluated across observation modes, communication constraints, sensor noise and team-size transfer.

  First author In preparation
• OCEANS 2026

  An adaptive open-source architecture for enhanced autonomy in BlueBoat ASVs based on ROS2

  The BlueBoat platform as a fully autonomous ocean-monitoring system — autonomy stack, onboard perception, mechanical and electronic integration, SITL simulation and field validation.

  Co-author Under review
• IEEE ICRA 2027

  Vision-based localization of small floating objects using ASVs with onboard cameras

  Hybrid synthetic-plus-real dataset feeding a YOLO detector, paired with a PPO policy with curiosity and LSTM memory trained in a purpose-built simulator for sim-to-real robustness. Currently in field-test phase.

  First author In preparation

Talks § transmissions

• 26 Mar 2026 Webinar · IEEE IMS Spanish Chapter

  Autonomous pop-up buoys for underwater observatories

  Invited speaker alongside Dr. Matias Carandell, Dr. Ivan Masmitja and Carlos de la Vega — on transmitting data from underwater observatories without physically recovering the instruments, and what that enables for long-term monitoring of deep marine ecosystems. See the announcement on LinkedIn.

Background § ship's papers

Beyond research § off watch

Contact § hailing channel

• Professional pedro.agundez@csic.es
• Personal pedroagundezfernandez@gmail.com
• LinkedIn Pedro Agundez Fernandez