Create an error chunk.

Build a JSON chat event frame. Includes runId in the payload to match the real Gateway protocol. Default run_id matches the default in _ack_r

Build a JSON agent event frame. Includes runId in the payload to match the real Gateway protocol.

Build a chat.send ACK response.

Clear buffered speaker reference and reset the APM state.

Collect all events from chat_send into a list.

Main event loop. Runs the state machine indefinitely until shutdown.

Start the worker thread that drives the 100 Hz control loop.

Dispatch a node.invoke.request to the appropriate handler.

Gracefully stop the event loop and close the WS.

Check if transcribed text contains a sleep phrase.

Build a minimal config dict for testing wake/sleep.

Create an APM echo canceller, skipping if the shim isn't available.

Connect to the Gateway and complete the protocol v3 handshake as a node.

Capture speech segment with pause detection. Returns: WAV bytes of captured speech, or None if no speech detected

Stop the active move and discard any queued primary moves. Thread-safe: executed by the worker thread via the command queue.

Get current face tracking offsets (thread-safe).

Process an arbitrary-length mic chunk through AEC. Returns cleaned PCM bytes (same length as input).

Transcribe WAV audio (16 kHz mono int16) and detect speech vs noise.

Extract text from a chat event payload.

Pick a random spontaneous speech prompt.

Connect to the Gateway and complete the protocol v3 handshake.

Sample the thinking animation at time *t* seconds.

Generate a sine wave as int16 PCM bytes.

Hermite smoothstep: 0→1 over t in [0,1].

Send a chat message and yield streaming events. Args: session_key: OpenClaw session key (e.g. "reachy") mess

Request the worker thread to stop and wait for it to exit. Before stopping, resets the robot to a neutral position.

Update the last activity time.

Create a client that's already 'connected' and return (client, ws).

Generate a synthetic WAV file as bytes.

Create a deep copy of a full body pose tuple.

Feed raw 16kHz int16 mono PCM from the speaker path.

Open a Dialog from a SolveIt dialog URL.

Get the latest frame (thread-safe). Returns a copy in BGR format.

Reset VAD internal state.

Enable or disable listening mode without touching shared state directly. While listening: - Antenna positions are frozen at the last

Set the user message for the next chat.send call.

Add nvidia pip-package library dirs to LD_LIBRARY_PATH at runtime. onnxruntime-gpu needs libcublas, libcudnn etc. When installed via pip (nv

Sleep in small increments so stop() takes effect quickly.

Determine if audio contains actual speech vs background noise. Uses multiple heuristics: - avg_logprob: How confident the mo

Compute next randomized delay (seconds).

Helper to transition back to sleep state.

Select the best face based on confidence and area. Uses the same weighted scoring as the YOLO tracker: score = confidence * 0.7 + nor

Set up OpenWakeWord model. Supports both explicit .onnx file paths and pretrained model names (e.g. "hey_jarvis_v0.1", "alex

Hermite smoothstep for jump-free transitions.

Inject a one-time assistant message into the session transcript. This enters model context on subsequent turns without running the agent.

Create a gateway client from the scaffold config dict. Reads openclaw config from config.yaml's openclaw section, falling back to enviro

Get head position from face detection. Same interface as YOLO HeadTracker. Args: img: Input image (BGR, from camera).

Queue a primary move to run after the currently executing one. Thread-safe: the move is enqueued via the worker command queue so the

Reset capture state for new speech segment.

Wake up animation. Uses robot's built-in wake_up() method.

Build the streaming middleware pipeline. Pipeline order (outermost to innermost): TTS → Filter → Sentence → Tool → Provider

Wrap PCM int16 bytes in a WAV container.

Manage circular buffer and detect voice activation. Args: chunk: Audio chunk vad_confidence: VAD result

Convert float32 numpy audio to WAV bytes (16-bit PCM mono).

Wrap raw PCM bytes in a WAV container (mono).

Process audio after activation, detect pauses. Args: chunk: Audio chunk vad_confidence: VAD result

Convert WAV bytes to float32 numpy array at 16kHz mono. Handles stereo→mono conversion and sample rate conversion if needed.

Capture current camera frame as an Anthropic-compatible image content block. Args: camera_worker: CameraWorker instance with get_latest_f

Encode a BGR numpy frame as an Anthropic image content block. Args: frame: BGR uint8 numpy array from camera. max_size: Max dimen

Connect if not already connected, or reconnect if the WS died.

Format a 6-tuple of offsets for display.

Identify the speaker from WAV audio bytes. Args: wav_bytes: WAV-format audio (16-bit PCM, any sample rate, mono or stereo).

Convert a Message to a plain dict, omitting empty fields.

Poll and log offsets for a duration. Args: cw: CameraWorker mm: MovementManager duration_s: how long to poll labe

Process stream through all middlewares in order. Args: stream: Input stream Yields: Processed chunk

Main loop: sleep, check conditions, perform gesture.

Start the camera worker loop in a thread.

Stop the camera worker loop.

Stop antenna wiggle animation and return to default neutral pose (singleton). Signals the wiggle loop to stop and waits for it to co

Build the prefixed message for chat.send. Prepends a [speaker:name] tag (or [speaker:unknown]) so the agent knows who is speaking. V

Get duration of WAV audio in seconds. Args: audio_wav: WAV audio bytes Returns: Duration in seconds

Create a temporary enrollments JSON file.

Set up mock emotions library on a manager.