Summary:
UCLA researchers in the Department of Mechanical and Aerospace Engineering have developed a waterproof and durable vision-based tactile sensor that can interrupt and replan action sequences when changes occur in the observation space. This design improves tactile sensing accuracy and robustness for robotic systems operating in harsh, high-wear, underwater, or contact-rich environments.
Background:
Vision-based tactile sensors are increasingly used in robotics because they are relatively low-cost, easy to manufacture, and adaptable across a wide range of form factors and manipulation tasks. These sensors typically include a deformable contact surface, an internal camera, and a light source. When the sensor contacts an object, the deformable surface changes shape, and the camera captures visual changes that can be analyzed to estimate contact forces, deformation, shear, or other tactile information. Despite their promise, many vision-based tactile sensors remain limited by durability. The deformable surfaces are often made from elastomeric materials that can tear, cut, abrade, or delaminate after repeated use. Replaceable contact surfaces can address wear, but they may reduce sensing accuracy or require repeated recalibration. Another challenge for these sensors is detecting and responding to tactile events. Diffusion policies are a powerful approach for adjusting action sequences in response to tactile events, but traditional diffusion policies operate with fixed action execution horizons, which make them incapable of responding to changes in the observational space during a fixed inference interval. A more robust and responsive tactile sensing architecture is needed to enable reliable robotic manipulation in challenging real-world environments.
Innovation:
UCLA researchers have developed a vision-based tactile sensor architecture that protects the deformable surface from mechanical damage and enables action sequences to be rapidly interrupted and replanned when changes are detected in the observation space. The sensor can be tuned for abrasion resistance, cut resistance, friction, shear sensitivity, electrostatic properties, or other task-specific mechanical characteristics. The technology is particularly well suited for robotic systems that must interact with abrasive, sharp, wet, or hard-to-access environments. In one implementation, the sensor is being developed as a robust underwater tactile sensor for use in robotic manipulation. The architecture can also be adapted to different form factors, including fingertip-style sensors, larger sensing surfaces, or multi-camera configurations, making it broadly applicable across robotic platforms.
Potential Applications:
● Robotic tactile sensing
● Dexterous robotic manipulation
● In-hand manipulation
● Pick-and-place robotics
● Contact-rich industrial automation
● Underwater remotely operated vehicles and manipulators
● Marine robotics and subsea inspection
● “Lights-out” manufacturing environments
● Handling of sharp, abrasive, or irregular objects
● Field-deployable robotic systems
● Search-and-rescue robotics
● Defense and naval robotic systems
● Warehouse and logistics automation
● Human-safe robotic grippers
● Robotic end-effectors requiring durable tactile feedback
Advantages:
● Improves durability of vision-based tactile sensors
● Recognizes and responds to temporally sparse tactile events
● Protects deformable sensing surfaces from abrasion, cutting, and wear
● Enables customizable tactile response through fabric pattern, texture, and material selection
● Supports sensing in harsh, high-cycle, underwater, or hard-to-maintain environments
● May reduce the need for frequent replacement or recalibration of sensing surfaces
● Compatible with multiple form factors and robotic end-effector designs
● Can be adapted for directional sensitivity, shear detection, friction tuning, or task-specific surface properties
● Enables tactile sensing for robotic handling of objects that may be unsafe or impractical for humans to manipulate directly
Development-To-Date:
First successful demonstration of the invention.
Related Papers:
• Scalable fabric tactile sensor arrays for soft bodies
Reference:
UCLA Case No. 2026-091
Lead Inventors:
Benjamin Forbes, Evan Harber, Veronica Santos