Reachy Mini Skill | Agent Skills

Reachy Mini

Control Reachy Mini robots using the Python SDK. Supports both real hardware and MuJoCo simulation.

Quick Start

from reachy_mini import ReachyMini
from reachy_mini.utils import create_head_pose
import numpy as np

with ReachyMini() as mini:
    # Smooth movement
    mini.goto_target(
        head=create_head_pose(pitch=15, degrees=True),  # Look up
        antennas=np.deg2rad([30, 30]),                  # Antennas out
        duration=1.0
    )

Connection Options

| Scenario | Code | |----------|------| | Local wired | ReachyMini() | | Wireless | ReachyMini(localhost_only=False) | | No camera/audio | ReachyMini(media_backend="no_media") | | Simulation | Start daemon with --sim, then ReachyMini() |

Movement

Head Control

Head pose is a 4x4 transformation matrix. Use create_head_pose() helper:

from reachy_mini.utils import create_head_pose

# Parameters: x, y, z (position), roll, pitch, yaw (rotation)
pose = create_head_pose(
    z=10, pitch=15,     # Up 10mm, pitch 15 degrees
    mm=True,            # Position in mm (default: meters)
    degrees=True        # Angles in degrees (default: True)
)
mini.goto_target(head=pose, duration=1.0)

Antennas

Two antennas controlled by angles in radians: [right, left]

mini.goto_target(antennas=np.deg2rad([45, -45]), duration=0.5)

Movement Methods

| Method | Use Case | |--------|----------| | goto_target() | Smooth interpolated movement | | set_target() | Instant position (high-frequency control) |

Interpolation methods: linear, minjerk (default), ease, cartoon

Built-in Behaviors

mini.wake_up()      # Init pose + sound
mini.goto_sleep()   # Sleep pose + sound
mini.look_at_world(x, y, z, duration=1.0)  # Look at 3D point
mini.look_at_image(u, v, duration=1.0)     # Look at pixel coordinate

Camera & Audio

Access via mini.media:

# Camera (returns BGR numpy array)
frame = mini.media.get_frame()

# Play sound (bundled or path)
mini.media.play_sound("wake_up.wav")

# Record audio
mini.media.start_recording()
sample = mini.media.get_audio_sample()  # Returns bytes or numpy array
mini.media.stop_recording()

# Stream audio output
mini.media.start_playing()
mini.media.push_audio_sample(audio_data)  # numpy float32 array
mini.media.stop_playing()

Building Apps

Subclass ReachyMiniApp to create installable apps:

from reachy_mini import ReachyMini
from reachy_mini.apps import ReachyMiniApp
import threading

class MyApp(ReachyMiniApp):
    custom_app_url = "http://0.0.0.0:8042"  # Optional web UI

    def run(self, reachy_mini: ReachyMini, stop_event: threading.Event):
        while not stop_event.is_set():
            # App logic here
            pass

For detailed app development: See references/app-development.md

LLM Integration Pattern

For voice/chat apps with tool calling:

# Define tools the LLM can call
tools = [
    {"name": "move_head", "params": {"direction": "left|right|up|down|center"}},
    {"name": "express_emotion", "params": {"emotion": "happy|sad|surprised"}}
]

# Execute tool calls
async def handle_tool(name, args):
    if name == "move_head":
        direction = args["direction"]
        poses = {
            "left": create_head_pose(yaw=20, degrees=True),
            "right": create_head_pose(yaw=-20, degrees=True),
            "up": create_head_pose(pitch=15, degrees=True),
            "down": create_head_pose(pitch=-15, degrees=True),
            "center": create_head_pose()
        }
        mini.goto_target(head=poses[direction], duration=0.5)

Resources

references/api.md - Complete API reference
references/app-development.md - App development guide

Agent Skills: Reachy Mini

Install this agent skill to your local

Skill Files