Getting your Trinity Audio player ready...
|
Robots often face challenges when planning and executing complex full-body manipulations. Performing intricate activities can be particularly difficult for them. Despite these challenges, there have been significant advancements in robotics, and researchers continue to work on improving robots’ capabilities for full-body manipulations and intricate activities.
Researchers from MIT have created an elongated, curved touch sensor with a human finger-like shape, which relies on cameras for tactile sensing. This innovative GelSight Svelte device offers high-resolution tactile feedback across a substantial surface area. Its unique design employs two mirrors to redirect and manipulate light, enabling a single camera positioned at the sensor’s base to capture visual data along the entire length of the finger-shaped sensor.
Furthermore, the scientists constructed the finger-shaped sensor with a flexible support structure. By monitoring the deformation of this support structure when the finger-shaped sensor comes into contact with an object, it can gauge the applied force.
They applied GelSight Svelte sensors to create a robotic hand capable of gripping heavy objects in a human-like manner, utilising the complete sensing area across all three of its fingers. This robotic hand was also proficient in executing typical pinch grasps commonly associated with traditional robotic grippers.
The constraints of conventional tactile sensors, relying on cameras, stem from their limited size, lens focal distances, and viewing angles. Consequently, these sensors are typically small and flat, confined to the fingertips of robots.
However, achieving an extended sensing area resembling a human finger poses challenges. The camera must be positioned at a considerable distance from the sensing surface to encompass such an area. This is incredibly complex due to robotic grippers’ size and shape limitations. To address this challenge, Zhao and Adelson devised an innovative solution. They employed a pair of mirrors manipulating light, directing it towards a single camera at the finger’s base.
The GelSight Svelte sensor incorporates two critical mirrors: one flat, angled mirror opposite the camera and another long, curved mirror positioned along the back of the sensor. These mirrors manipulate the light rays from the camera in a manner that allows it to capture the entire length of the finger’s surface.
To optimise these mirrors’ shape, angle, and curvature, the researchers developed specialised software capable of simulating light reflection and refraction. Zhao elaborated, “With this software, we can easily experiment with the placement and curvature of the mirrors to predict the quality of the image after constructing the sensor.”
These mirrors, cameras, and dual sets of LEDs for illumination are affixed to a plastic backbone and enveloped in a flexible silicone gel skin. The camera observes the skin’s interior; monitoring its deformations can detect contact points and assess the geometry of the object’s contact surface.
Moreover, the red and green LED arrays provide insight into the depth of the gel’s compression when an object is grasped, as indicated by variations in colour saturation at different sensor locations. This colour saturation data enables the researchers to reconstruct a 3D depth image of the grasped object.
The sensor’s plastic backbone is crucial in determining proprioceptive information, including twisting torques applied to the finger. When an object is grasped, the spine bends and flexes. Through machine learning algorithms, the researchers can estimate the applied force on the sensor based on these backbone deformations.