Embodied AI Glossary中文

Contact Microphone

接触式麦克风(音频触觉)Advanced

A microphone stuck to a gripper or object that picks up contact vibration, used as a cheap tactile sensor.

A contact microphone doesn’t pick up sound through the air — it’s attached to a solid surface and picks up vibration directly, most commonly using a piezoelectric element. In robotics it’s used as a cheap, durable substitute for touch: the high-frequency vibrations from scraping, collision, friction, or shaking can reveal whether contact has occurred, what material a surface is, or whether something is inside a container — things a camera often can’t see. Hearing Touch (ICRA 2024) uses a contact microphone as a stand-in for touch and improves a manipulation policy with representations learned from large-scale audio-visual pretraining; ManiWAV (CoRL 2024) embeds a piezoelectric contact microphone in one finger of a UMI handheld gripper, so collected human demonstrations carry both video and sound. Its limitation is that it can’t help when the interaction itself makes no sound.

ExampleIn ManiWAV’s dice-pouring task, the robot first shakes a cup and listens through the contact microphone for the vibration of dice hitting the cup, judging whether the cup has been fully emptied.

Also called
Audio as Tactile Signal, Piezo Contact Microphone
Related
Tactile Sensor · Multimodal Perception · Universal Manipulation Interface · Contact-rich Manipulation · Visuo-Tactile Fusion · Microphone Array
Sources
Hearing Touch: Audio-Visual Pretraining for Contact-Rich Manipulation (arXiv 2405.08576)
ManiWAV: Learning Robot Manipulation from In-the-Wild Audio-Visual Data (arXiv 2406.19464)

See it in the full glossary →