Researchers at Columbia Engineering University in New York have produced a robotic hand that can carry out tasks without visual input from cameras. The researchers created the dexterous robotic hand with five fingers that have 15 joints that can operate independently of one another. By using electric motors to control the fingers and special tactile sensors equipped with reinforcement learning techniques, the robotic hand is able to carry out tasks using only the sense of touch. Scientists trained the hand’s manipulation skills using deep reinforcement learning in a simulator, using only tactile and proprioceptive data. The researchers think that their proof-of-concept robotic hand could become useful for logistics, material transport, and product manufacture.
