UTact: Underwater Vision‐Based Tactile Sensor with Geometry Reconstruction and Contact Force Estimation

Abstract

Soft robots offer potential advantages for underwater applications such as environmental exploration and delicate manipulation, thanks to their inherent flexibility and sustainability. Embedded sensing systems further enhance these capabilities by providing environmental information feedback. Until now, many flexible sensors integrated into soft robots suffer from slow response times and limited sensing stability. Vision‐based tactile sensing presents a promising solution to these challenges, yet its application underwater remains limited. In this work, we introduce a compact, reliable, and highly sensitive underwater vision‐based tactile sensor (UTact) featuring a simple fabrication process and low cost. UTact enables 3D reconstruction of the tactile surface, complemented by image enhancement algorithms optimized for underwater environments. A ResNet‐34 model is trained and validated for underwater sediment classification, demonstrating high accuracy. Additionally, the sensor's ability to estimate contact force during pressing on several common objects is evaluated and verified. This research underscores the potential of vision‐based tactile sensors for underwater soft robotics and suggests promising avenues for future integration into underwater robotic systems.