Embodied Passive Aeroacoustic Perception Enables Relative Sensing and Pursuit Between Aerial Robots

Yanbaihui Liu, Ravi Prakash, Li-Yu Lo, Nils Roede, Boyuan Chen

Duke University

website | paper | video

Overview

Aerial robots continuously generate structured aeroacoustic fields during flight, yet these signals have been largely underexplored as a source of onboard relative perception, particularly under the strong ego-acoustic interference generated during simultaneous flight in various outdoor conditions. We introduce \textit{embodied passive aeroacoustic perception}, a sensing paradigm in which an aerial robot infers actionable relative-state information from the naturally generated sound of flight while operating within its own evolving aeroacoustic field. We present SonicFly, a passive aeroacoustic perception framework that enables one unmanned aerial vehicle (UAV) to estimate and follow another using only the leader's intrinsic flight sound, without active acoustic signaling, inter-robot communication, GPS sharing, or external sensing infrastructure. The system uses a lightweight four-microphone array, rotorcraft-informed acoustic representations, a neural bearing-range estimator, and confidence-gated filtering for closed-loop flight. Through acoustic characterization, onboard localization, and outdoor pursuit experiments, we show that multirotor aeroacoustic signals contain sufficient information to support relative perception despite strong ego-acoustic interference, environmental variability, and continuously changing flight geometry. During acoustic-only pursuit, SonicFly achieved a mean distance-maintenance error of 1.34 m across diverse outdoor trajectories and operating conditions. Analysis of the acoustic channel further reveals design principles governing embodied passive aeroacoustic perception, including the roles of harmonic structure, spectral separability, and spatial acoustic cues in determining observability. Our results establish the feasibility of embodied passive aeroacoustic perception for aerial robots and suggest that naturally generated behavioral signals can serve as useful information for robotic perception and coordination.

Repository Layout

.
├── main.py                         # Train SonicScoutNet
├── evaluate.py                     # Evaluate a trained checkpoint
├── requirements.txt                # Python dependencies
├── pretrained/
│   ├── sonic_scout_net_sincos_sonic.pth
│   └── normalization_params_sonic.json
├── src/
│   ├── sonicscout_model.py         # CNN model architecture
│   ├── shared_dataset.py           # Dataset loader and metrics
│   ├── confidence_utils.py         # Deployment confidence scoring
│   └── kf.py                       # Kalman filtering utilities
└── postprocessing/                 # ROS bag/audio/log/data conversion scripts

Prerequisites

1. Clone the repository:

   git clone https://github.com/generalroboticslab/SonicFly.git

2. Set up the environment:

    python3 -m venv .venv
     source .venv/bin/activate
     pip install -r requirements.txt

Dataset

Download the prepared dataset from the Google Drive folder. Create a dataset/ directory in the repository root and place the files directly inside it:

SonicFly/
└── dataset/
    ├── trainval.npz
    ├── test.npz
    ├── audio_cmvn.npz
    └── normalization_params.json

Training

python main.py

Resume training with:

python main.py --resume pretrained/sonic_scout_net_sincos_sonic.pth

Evaluation

python evaluate.py

Results are written to evaluation_results_sonic/, including metrics, predictions, and evaluation figures. Evaluation uses the deployment confidence scoring method by default.

Bearing uses the body-frame convention [-180, 180]; positive values indicate the leader is to the follower's left.

BibTeX

If you find our paper or codebase helpful, please consider citing:

@misc{liu2026embodiedpassiveaeroacousticperception,
      title={Embodied Passive Aeroacoustic Perception Enables Relative Sensing and Pursuit Between Aerial Robots}, 
      author={Yanbaihui Liu and Ravi Prakash and Li-Yu Lo and Nils Roede and Boyuan Chen},
      year={2026},
      eprint={2608.00401},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2608.00401}, 
}

Acknowledgement

Research was sponsored by the U.S. Army Combat Capabilities Development Command Army Research Laboratory (ARL) and was accomplished under Contract No. W911QX23C0011. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of ARL. The U.S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation herein. This work was also supported by DARPA TIAMAT program under award HR00112490419 and ARO under award W911NF2410405.