With the help of AI and a neural network, this wiggling robotic finger digs into substances such as sand to locate and identify buried objects. Robotic hands, fingers, grippers, and their various sensors are usually deployed to function in “free space” for many obvious reasons. But in the real world, it’s often necessary to poke and probe into materials such as sand while looking for objects like buried nails or coins (best case) or explosives (a far-worse case).
Addressing this challenge, a team at the Massachusetts Institute of Technology (MIT) has devised and tested their second-generation robotic finger specifically designed for this class of mission with impressive success: Their “Digger Finger” was able to dig through granular media such as sand and rice and correctly sense the shapes of submerged items it encountered. The objective is an electromechanical finger that replicates the capabilities of a human one, which can find, feel, and “see” (mentally) the shape of a buried object based on sensing and learned experience.
The project consists of two major aspects: the robotic finger that can penetrate into a granular mass to get close enough to illuminate and capture a reflected—albeit highly distorted—pattern, followed by artificial intelligence/neural-net data-extraction and analysis algorithms. At the core of the hardware design is GelSight, a sharp-tipped robot finger developed by researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL).
What you’ll learn:
- Why and how a complicated robotic-finger design was improved by a research team.
- How this finger is used to dig into bulk materials to locate and “see” small objects.
- How the reflected image of the object is the basis for the data set that determines the actual shape of the object.
- How AI and a neural net were employed to analyze the distorted, reflected object images that were captured.