Semantic Audio-Visual Navigation
This work addresses the limitation of existing audio-visual navigation systems that assume continuous sound, making it more applicable to real-world scenarios for robotic navigation.
This paper introduces semantic audio-visual navigation, where agents must find objects based on sporadic, semantically meaningful sounds. The proposed transformer-based model, using a multimodal memory and an inferred goal descriptor, significantly outperforms existing audio-visual navigation methods in this new task.
Recent work on audio-visual navigation assumes a constantly-sounding target and restricts the role of audio to signaling the target's position. We introduce semantic audio-visual navigation, where objects in the environment make sounds consistent with their semantic meaning (e.g., toilet flushing, door creaking) and acoustic events are sporadic or short in duration. We propose a transformer-based model to tackle this new semantic AudioGoal task, incorporating an inferred goal descriptor that captures both spatial and semantic properties of the target. Our model's persistent multimodal memory enables it to reach the goal even long after the acoustic event stops. In support of the new task, we also expand the SoundSpaces audio simulations to provide semantically grounded sounds for an array of objects in Matterport3D. Our method strongly outperforms existing audio-visual navigation methods by learning to associate semantic, acoustic, and visual cues.