Researchers at MIT have developed a robotic hand, named GelSight EndoFlex, that can identify objects with a single grasp by using high-resolution touch sensors and machine learning. The hand has three fingers and is constructed with a soft silicone skin enclosed within a rigid skeleton, which contains small, spherical sensors that detect the shape and texture of objects.
Once the hand grasps an object, the sensors transmit data to a machine learning algorithm, which uses this information to identify the object with an 85% accuracy rate. The team anticipates that the hand could have numerous applications in areas such as manufacturing and healthcare. In manufacturing, the hand could be utilized to inspect products for defects, while in healthcare, it could assist with surgical procedures.
The hand is designed to allow it to identify a wide range of objects with various shapes, sizes, and textures. The hand’s sensors are also capable of detecting minute changes in surface texture and shape, which enables it to identify objects that are similar in appearance but different in texture.
Although the hand is still in its preliminary development stages, the researchers predict that it could significantly transform the way that robots engage with the environment.