Swoosh! Rattle! Thump! -- Actions that Sound

TL;DR

This paper introduces a large-scale dataset capturing sound, action, and vision interactions in robotics, revealing that sound provides valuable information for object identification, causal inference, and physical property prediction.

Contribution

It presents the first extensive dataset linking sound and robotic actions, and demonstrates how audio embeddings enhance physical understanding and prediction in robotics.

Findings

Sound differentiates object classes like screwdrivers and wrenches.

Audio data enables predicting applied actions from sound.

Audio embeddings outperform visual embeddings in predicting physical properties.

Abstract

Truly intelligent agents need to capture the interplay of all their senses to build a rich physical understanding of their world. In robotics, we have seen tremendous progress in using visual and tactile perception; however, we have often ignored a key sense: sound. This is primarily due to the lack of data that captures the interplay of action and sound. In this work, we perform the first large-scale study of the interactions between sound and robotic action. To do this, we create the largest available sound-action-vision dataset with 15,000 interactions on 60 objects using our robotic platform Tilt-Bot. By tilting objects and allowing them to crash into the walls of a robotic tray, we collect rich four-channel audio information. Using this data, we explore the synergies between sound and action and present three key insights. First, sound is indicative of fine-grained object class information, e.g., sound can differentiate a metal screwdriver from a metal wrench. Second, sound also contains information about the causal effects of an action, i.e. given the sound produced, we can predict what action was applied to the object. Finally, object representations derived from audio embeddings are indicative of implicit physical properties. We demonstrate that on previously unseen objects, audio embeddings generated through interactions can predict forward models 24% better than passive visual embeddings. Project videos and data are at https://dhiraj100892.github.io/swoosh/