That Sounds Right: Auditory Self-Supervision for Dynamic Robot Manipulation - Technology Org

When humans try to strike a tennis ball, unlock a door, or crack an egg, they rely on sound as a signal of success. Similarly, using auditory signals can help robots learn motor skills. AURL learns to generate dynamic behaviors from contact sounds produced from a UR10 robot's interaction. It uses multi-channel sound data to output parameters that control dynamic behavior. To learn low-dimensional representations from audio, self-supervised learning methods are used.