Currently, human-like risk perception is missing in decision-making and control in autonomous systems.

Developing autonomous marine systems that can thoroughly assess and model risks would enable them to behave more intelligently, interact safely in unstructured environments and foster public trust. 

The ERC-funded BREACH project will tackle this challenge by focusing on risk-based autonomy and shared control with human supervisors. 

Objective

Autonomous systems able to comprehensively assess and model risks would be a scientific breakthrough leading to highly intelligent system behaviors, safe interactions in unstructured environments, and public trust. 

Currently, risk perception is an important part of human cognition not adequately implemented in the situation awareness, control and decision-making of autonomous systems. 

Therefore, this project addresses fundamental research challenges related to risk-based autonomy and safe and efficient shared control with human supervisors. 

The goal is to create powerful risk modelling and risk control solutions, closely connected to risk-based decision support for human supervisors. 

Although there are a few preliminary concepts coupling risk models with control algorithms and artificial intelligence (AI), a fundamental framework providing risk-based control and decision support – for both the autonomous system and human supervisors – does not exist. 

A new framework

BREACH will establish a framework constituted by quantitative risk models, integrating hardware, software, human, and environmental risk elements. 

The complexity of the framework will gradually evolve in two parallel tracks along with demonstrations for mission planning and execution for autonomous marine “organizations”. 

The focus on marine systems is due to the demanding challenges in that domain, and my knowledge and experience as a researcher in the Centre of Excellence on Autonomous Marine Operations and Systems (NTNU AMOS). 

Even though the focus is on marine autonomy, the results should be applicable to other application areas.

Principal Investigator

• Professor Ingrid Bouwer Utne

Funding and duration

The project is funded by the The European Research Council ERC Advanced Grant. Duration of the project is from 1 December 2025 – 31 December 2029. Grant agreement ID: 101142277