Modular Collision Avoidance Using Predictive Safety Filters

Abstract

The number of maritime projects is increasing yearly, including offshore applications, underwater robotics for ocean condition monitoring, and autonomous ship transport. Many of these activities are safety-critical, making it essential to have a robust closed-loop control system that satisfies constraints arising from underlying physical limitations and safety aspects. However, this is often challenging to achieve for real-world systems. For example, autonomous ships at sea have non-linear and uncertain dynamics and are subject to numerous time-varying environmental disturbances such as waves, currents, and wind. There is growing interest in using machine learning-based approaches to adapt these systems to more complex scenarios. However, there is currently no standard framework to guarantee the safety and stability of such systems. Predictive safety filters have emerged recently as a valuable method for ensuring constraint satisfaction, even when unsafe control inputs are used. The safety filter approach leads to a modular separation of the problem, allowing the usage of arbitrary control policies in a task-agnostic way. In this work, a predictive safety filter is developed to ensure anti-grounding and ship collision avoidance for a small prototype ferry. The filter takes in a nominal input sequence from a potentially unsafe controller and solves an optimization problem to compute a minimal perturbation of the nominal control inputs, which adheres to physical and safety-related constraints. The system is validated by simulations for several realistic scenarios with map data from Trondheim, Norway. It is demonstrated that the predictive safety filter can avoid collisions with static and dynamic obstacles. The predictive safety filter approach is flexible and can be used to improve the robustness of various offshore applications, e.g. wind turbine stabilization, autonomous vessels, and marine robotics.