Sound Direction of Arrival (DoA)

This example demonstrates how to use the microphone array to detect the Direction of Arrival (DoA) of speech. The robot uses the FastAPI endpoint to get DoA information, calculates the position of the sound source, transforms it into world coordinates, and automatically looks towards the speaker.

How it works:

Continuously polls the /api/state/doa endpoint to get speech direction
When speech is detected, calculates the 3D position of the sound source
Transforms the position from head coordinates to world coordinates
Commands the robot to look at the speaker using look_at_world()

Features:

Automatic detection of robot IP (local or wireless)
Threshold-based filtering to avoid excessive head movements
Real-time transformation from head to world coordinates


import time

import numpy as np
import requests

from reachy_mini import ReachyMini


def main() -> None:
    """Continuously monitor audio input and orient the head toward the speaker."""
    with ReachyMini(log_level="DEBUG", automatic_body_yaw=True) as mini:
        # Get the robot IP address
        robot_ip = "localhost"
        if mini.client.get_status()["wireless_version"]:
            robot_ip = mini.client.get_status()["wlan_ip"]

        doa_url = f"http://{robot_ip}:8000/api/state/doa"

        last_doa = -1
        THRESHOLD = 0.004  # ~2 degrees

        while True:
            # Get DoA from FastAPI endpoint
            try:
                response = requests.get(doa_url, timeout=1.0)
                response.raise_for_status()
                doa_data = response.json()

                if doa_data is None:
                    print("  No DoA data available (audio not initialized)")
                    time.sleep(0.5)
                    continue

                angle = doa_data["angle"]
                speech_detected = doa_data["speech_detected"]

                print(f"DOA: angle={angle:.3f} rad, speech_detected={speech_detected}")

                if speech_detected and np.abs(angle - last_doa) > THRESHOLD:
                    print(f"  Speech detected at {angle:.1f} radians")
                    p_head = [np.sin(angle), np.cos(angle), 0.0]
                    print(
                        f"  Pointing to x={p_head[0]:.2f}, y={p_head[1]:.2f}, z={p_head[2]:.2f}"
                    )
                    T_world_head = mini.get_current_head_pose()
                    R_world_head = T_world_head[:3, :3]
                    p_world = R_world_head @ p_head
                    print(
                        f"  In world coordinates: x={p_world[0]:.2f}, y={p_world[1]:.2f}, z={p_world[2]:.2f}"
                    )
                    mini.look_at_world(p_world[0], p_world[1], p_world[2], duration=0.5)
                    last_doa = angle
                else:
                    if not speech_detected:
                        print("  No speech detected")
                    else:
                        print(
                            f"  Small change in DOA: {angle:.1f} rad (last was {last_doa:.1f} rad). Not moving."
                        )
                    time.sleep(0.5)

            except requests.RequestException as e:
                print(f"  Error fetching DoA: {e}")
                time.sleep(0.5)


if __name__ == "__main__":
    try:
        main()
    except KeyboardInterrupt:
        print("Exiting...")

Update on GitHub

Reachy Mini

Sound Direction of Arrival (DoA)