AI for Good Webinar - Robots in the wild: Decision-making AI for environmental monitoring

One of our research streams in GOALS is the development and application of decision-making under uncertainty models for robotic monitoring of the natural environment. This includes our work on ocean sampling using gliders (e.g., the MAS-DT project) and on biodiversity change assessment in grasslands using mobile robots (e.g., this paper).  

On January 26th 2026, Nick will present a high-level overview of the technical underpinnings of this work stream in the AI for Good - Discovery: AI and Robotics webinar programme. You can sign up for the webinar on the  event page.

The abstract is below:

Autonomous robots could transform environmental monitoring by collecting long-term, large-scale data at spatial and temporal resolutions that manual methods can’t match, whilst also offering lower carbon cost and greater adaptability. Yet routine deployment “in the wild” remains rare. Real-world monitoring demands robust hardware and, crucially, AI control that can make reliable tactical and strategic decisions under uncertainty despite dynamic conditions, limited sensing, and mission risk.

This talk presents an approach to decision-making under uncertainty for autonomous environmental monitoring missions. The robot-environment system is modelled as a Markov decision process (MDP), enabling principled planning and adaptation when outcomes are stochastic and information is incomplete. The talk shows how this framework supports reliable autonomy across diverse domains and platforms: dive planning for ocean gliders, demonstrated in over 1500 km of autonomous sampling in the North Atlantic; autonomous navigation strategies for biodiversity estimation in grassland ecosystems; and autonomous exploration and mapping for complex nuclear decommissioning environments.