MIT's Sensor-Packed Glove Helps Neural Networks Learn
MIT has developed a sensor-packed glove that users wear while handling a variety of objects. The sensors inside the glove have allowed the researchers to compile a large data set that is used to help an AI system recognize objects through touch alone. Researchers say the information gathered with the glove could help robots to identify and manipulate objects and could aid in prosthetic design.
The glove is a low-cost knitted unit that is called the "scalable tactile glove" or STAG and has about 550 tiny sensors across the entire hand. Each of those tiny sensors captures pressure signals as humans interact with objects in various ways. The data is sent to a neural network that learns a dataset of pressure-signal patterns related to specific objects.
The system then uses the dataset to classify the objects and predict the weight by feel alone with no visual input needed. There are similar sensor gloves on the market today that run into the thousands of dollar range, the MIT glove was made from commercially available materials and cost about $10. Those more expensive gloves also have only about 50 sensors and capture much less information.
MIT researchers say that the tactile sensing system could be used in combination with traditional computer vision and image-based datasets to give the robots more human-like understanding of interacting with objects. The highly sensitive nature of the MIT glove allows the researchers to use the data set to measure cooperation between regions of the hand during object interactions.
The scientists think that prosthetics manufacturers can potentially use the information gathered by the glove to choose optimal spots for sensor placement to help customize prosthetics to tasks and the objects the user interacts with regularly. The STAG glove is laminated with an electrically conductive polymer that changes resistance to apply pressure, and it features conductive threads that go through holes in the polymer from fingertips to the base of the palm that overlap in a way to turn them into pressure sensors.