Research · JSAI SIG-Challenge 2024 · 2024-07
JSAI SIGAI 2024: Sound Source Tracking with World Models
Developed a robot with a microphone array that uses a DreamerV3 world model to track sound sources in vision-denied scenarios.
Overview This research focused on situations where vision fails, such as finding objects in the dark. I developed a robot with a microphone array to track a sound source. Key Achievements Implemented the DreamerV3 world model based RL method. The agent learned the temporal and spatial dynamics of the sound's direction, allowing it to build an internal model of the acoustic environment. This internal model enabled the agent to learn efficient, long term exploration strategies. YouTube Demo Paper: https://www.jstage.jst.go.jp/article/jsaisigtwo/2024/Challenge 066/2024 07/ article/ char/ja/
#Robotics #RL #World Models #DreamerV3 #Acoustics