MARL-AUV for Spatiotemporal Mapping

ICRA-25 submission in preparation

Autonomous salinity mapping is vital for being able to understand estuarine dynamics and environmental change. Our work uses a multi-agent Dueling DQN framework to train AUVs for efficient, reward-driven sampling in a dynamic river plume. Agents observe local salinity, wind, and flow vectors to balance exploration with energy constraints, learning policies that minimize map error against long-term ground truth data. Research is ongoing and will be available soon on arXiv—code available upon request.

📽️ Watch 3 agents in action!