Assessing Vision-Language Models for Perception in Autonomous Underwater Robotic Software

A new study assesses Vision-Language Models (VLMs) for perception in Autonomous Underwater Robots (AURs), focusing on detecting underwater trash. VLMs show promise by generalizing to unseen objects and remaining robust in noisy, low-visibility conditions where traditional deep learning struggles. The 10-page empirical evaluation, conducted for an industrial maritime partner, measures performance and uncertainty to help software engineers select trustworthy models for critical AUR software perception modules.