# Marcus — End-to-End Pipeline

**Robot persona:** Sanad (wake word + self-intro)
**Updated:** 2026-04-28

One map of every data path from sensor to motor, voice to speech. Cross-reference with `architecture.md` (what each file is), `functions.md` (exact function signatures — AST-generated), and `MARCUS_API.md` (usage examples + JSON schemas).

---

## Boot sequence

`Brain/marcus_brain.py::init_brain()` — called once from `run_marcus.py` or `marcus_server.py`.

```
run_marcus.py
      │
      ▼
init_brain()
      │
      ├─ init_zmq()                            PUB bind tcp://127.0.0.1:5556 → Holosoma
      ├─ start_camera()                        RealSense 424×240@15fps → shared _raw_frame
      ├─ init_yolo(raw_frame, raw_lock)        YOLOv8m CUDA FP16, 19 classes — background thread
      ├─ init_odometry()                       ROS2 /dog_odom → dead reckoning fallback
      ├─ init_memory()                         loads Data/Brain/Sessions/session_NNN/
      │
      ├─ if subsystems.lidar:       init_lidar()         multiprocessing spawn SLAM_worker
      ├─ if subsystems.imgsearch:   init_imgsearch()     (off by default)
      ├─ if subsystems.autonomous:  AutonomousMode()     patrol state machine
      │
      ├─ send_cmd("start") + 0.5s + send_cmd("walk") + 0.5s   Holosoma handshake
      │
      ├─ if subsystems.voice:       _init_voice()        ▼ voice pipeline below
      └─ _warmup_llava()                        first Qwen2.5-VL inference
                                                "SANAD AI BRAIN — READY"
```

Subsystem flags live in `config_Brain.json::subsystems`. Current defaults:

```json
"subsystems": { "lidar": true, "voice": true, "imgsearch": false, "autonomous": true }
```

---

## Voice pipeline (when `subsystems.voice = true`)

Marcus uses **Gemini Live STT-only** for the user's mic plus **G1 TtsMaker** for the brain's spoken reply. No local wake detector — Gemini's server-side VAD decides turn boundaries; the wake-word check happens at dispatch time on the transcribed text.

```
G1 body mic (array)
  └─ UDP multicast 239.168.123.161:5555 ── int16 mono 16 kHz PCM
        ▼
Voice/audio_io.py::BuiltinMic
  ring buffer (64 KB) + read_chunk(n)        (Sanad-pattern; see audio_io.py)
        ▼
Voice/gemini_script.py::GeminiBrain  (asyncio worker thread)
  ├─ client.aio.live.connect(model="gemini-2.5-flash-native-audio-preview-12-2025",
  │                          config=LiveConnectConfig(
  │                            response_modalities=["TEXT"],          ← STT-only
  │                            input_audio_transcription={},
  │                            realtime_input_config=AutomaticActivityDetection(
  │                              start_of_speech_sensitivity=HIGH,
  │                              end_of_speech_sensitivity=LOW,
  │                              prefix_padding_ms=20,
  │                              silence_duration_ms=200),
  │                            system_instruction=<transcriber-only role>))
  ├─ _send_mic_loop  → 512-sample PCM chunks (32 ms each) → session.send_realtime_input
  ├─ _receive_loop   → server_content.input_transcription.text → on_transcript + on_command
  └─ on turn_complete → recorder.finish_turn() → "listening" log
        ▼
Voice/marcus_voice.py::VoiceModule._dispatch_gemini_command(text, "en")
  ├─ 1. _has_wake_word(text)
  │        match any of stt.wake_words variants as a whole word — else return early
  ├─ 2. _strip_wake_word(text)
  │        iterative until stable, "Sanad. Sanad." → "" / "Sanad turn right" → "turn right"
  ├─ 3. garbage / min-length filter
  │        skip "okay"/"thanks"/single-letter unless command_vocab matches exactly
  ├─ 4. _normalize_command(stripped)
  │        difflib fuzzy-match vs stt.command_vocab
  │        "Turn right up" → "turn right"  (canonical form)
  ├─ 5. dedup vs last_gemini_canon within command_cooldown_sec
  └─ 6. on_command(text, "en")
        ▼
Brain/marcus_brain.py::_on_command  (closure inside init_voice)
  ├─ flush_mic()                              ← drop pending mic audio
  ├─ result = process_command(text)
  │     ├─ regex fast-path → Brain/command_parser.py::try_local_command()
  │     │    places · odometry walk/turn · patrol · session recall · goal_nav
  │     │    + SIMPLE_DIR ("go back", "right", "forward") · STOP_SIMPLE ("stop", "halt")
  │     │    + NAT_GOAL_RE (naturalised goals like "the chair") · auto on/off
  │     │    (~50 ms when matched — NO LLM call)
  │     ├─ _TALK_PATTERNS ("what / who / where / …") → _handle_talk(cmd)
  │     │    → API/llava_api.py::ask_talk(...) → Qwen2.5-VL
  │     └─ else → _handle_llava(text)
  │            ├─ get_frame()  (10×50 ms poll, no 1 s stall)
  │            ├─ API/llava_api.py::ask(text, img)
  │            │    ollama.chat(qwen2.5vl:3b, num_batch=128, num_ctx=2048, num_predict=120)
  │            │    → parse_json() → {actions, arm, speak, abort}
  │            └─ Brain/executor.py::execute(d)
  │                    ├─ actions → MOVE_MAP[dir] → API/zmq_api.py::send_vel → Holosoma
  │                    ├─ arm     → API/arm_api.py  (stub for now)
  │                    └─ abort   → gradual_stop()
  ├─ audio_api.speak(result["speak"])         ← TtsMaker via G1 firmware
  └─ flush_mic()                              ← drop the speaker's echo from mic buffer
        ▼
API/audio_api.py::speak(text, lang="en")
  ├─ Voice/builtin_tts.py::BuiltinTTS.speak(text)
  │    client.TtsMaker(text, speaker_id=0)   — G1 on-board engine, English only
  │    time.sleep(len(text) * 0.08)
  └─ → back to listening
```

**Config knobs** (all in `config_Voice.json::stt`):
- Gemini connection: `gemini_model`, `gemini_voice_name`, `gemini_audio_profile`, `gemini_chunk_size`, `gemini_send_sample_rate`
- Gemini VAD: `gemini_vad_start_sensitivity`, `gemini_vad_end_sensitivity`, `gemini_vad_prefix_padding_ms`, `gemini_vad_silence_duration_ms`
- Gemini session lifecycle: `gemini_session_timeout_sec`, `gemini_max_reconnect_delay_sec`, `gemini_max_consecutive_errors`, `gemini_no_messages_timeout_sec`
- Persona: `gemini_system_prompt` (inline) or `gemini_system_prompt_file` (path)
- Recording (debug WAVs): `gemini_record_enabled`
- Mic gain: `mic_gain`
- Dispatch: `wake_words` (gate), `command_vocab` (fuzzy-match target), `garbage_patterns`, `command_vocab_cutoff`, `min_transcription_length`, `command_cooldown_sec`
- Hardware: `mic_udp.{group,port,buffer_max_bytes,read_timeout_sec}`, `speaker.{dds_interface,volume,app_name,begin_stream_pause_sec,wait_finish_margin_sec}`

**Env overrides** (highest precedence): `MARCUS_GEMINI_API_KEY` (or `SANAD_GEMINI_API_KEY` fallback), `MARCUS_GEMINI_MODEL`, `MARCUS_GEMINI_VOICE`.

---

## Terminal / WebSocket command pipeline (same brain, skips voice)

```
run_marcus.py stdin   OR   Server/marcus_server.py WebSocket
        ▼
Brain/marcus_brain.py::process_command(text)
        ▼  (same parser → LLaVA → executor → ZMQ as above)
        ▼
result dict  →  stdout   OR   WebSocket reply frame
```

---

## Vision pipeline (continuous, consumed by brain on demand)

```
RealSense D435 (USB)
  └─ 424×240 BGR 15 fps
      → API/camera_api.py — shared _raw_frame (thread-safe)
                    │                  │
                    │                  └─ get_frame() → JPEG base64 on demand
                    ▼
       Vision/marcus_yolo.py (daemon thread)
       YOLOv8m @ cuda:0 FP16 imgsz=320
       → _latest_detections (thread-safe list)
         yolo_sees / yolo_closest / yolo_summary / yolo_fps
                    ▼
       Navigation/goal_nav.py  (fast YOLO check → Qwen-VL fallback)
       Autonomous/marcus_autonomous.py  (patrol scan every N steps)
       Brain/marcus_brain.py  (status / alerts)
```

---

## Movement pipeline

```
Brain/executor.py  OR  Brain/command_parser.py  OR  Navigation/*
        │   uses MOVE_MAP from config_Navigation.json
        ▼
API/zmq_api.py::send_vel(vx, vy, vyaw)  JSON over ZMQ PUB (port 5556)
        ▼
Holosoma RL policy (separate process, hsinference env)
        ▼
G1 low-level joint commands over DDS/eth0
        ▼
29-DOF body motion
```

---

## LiDAR pipeline (when `subsystems.lidar = true`)

```
Livox Mid-360 (192.168.123.120, UDP)
        ▼
Lidar/SLAM_worker.py  (multiprocessing.spawn subprocess — CUDA-safe spawn)
    ├─ SLAM_engine, SLAM_Filter, SLAM_LoopClosure, SLAM_Submap, SLAM_NavRuntime
    ├─ publishes pose + obstacle flags back to parent via Queue
    └─ writes occupancy grids to Data/Navigation/Maps/
        ▼
API/lidar_api.py  (reads the queues, exposes:)
        ├─ obstacle_ahead() → bool
        ├─ get_lidar_status() → dict (pose, loc_state, frame age, FPS, ICP ms)
        └─ LIDAR_AVAILABLE
        ▼
Navigation/goal_nav.py rotation thread — pauses motion on obstacle_ahead()
Brain/command_parser.py — responds to "lidar status" queries
```

---

## Knobs that control each stage

| Knob | Location | Effect |
|---|---|---|
| `subsystems.lidar` | config_Brain.json | SLAM subprocess on/off |
| `subsystems.voice` | config_Brain.json | Gemini Live STT + dispatch + TtsMaker loop on/off |
| `subsystems.imgsearch` | config_Brain.json | image-guided search init on/off |
| `subsystems.autonomous` | config_Brain.json | auto-patrol state machine init on/off |
| `num_batch`, `num_ctx` | config_Brain.json | llama.cpp compute-graph size (128 / 2048 ≈ 1.8 GiB graph — **do not raise** on 16 GB Jetson) |
| `num_predict_main` | config_Brain.json | 120 tokens max for the main JSON reply |
| `yolo_device`, `yolo_half` | config_Vision.json | `cuda` / FP16 (hard-required; CPU not allowed) |
| `mic.backend` | config_Voice.json | `builtin_udp` (G1 array — only option used by Gemini path) |
| `mic_udp.group/port` | config_Voice.json | where to join the G1 audio multicast |
| `mic_udp.read_timeout_sec` | config_Voice.json | `BuiltinMic.read_chunk` budget (default 0.04 s) |
| `tts.backend` | config_Voice.json | `builtin_ttsmaker` (only supported option) — used by `AudioAPI.speak()` for the brain's reply |
| `stt.wake_words` | config_Voice.json | 33 fuzzy variants of "Sanad" — wake-word gate at dispatch time |
| `stt.command_vocab` | config_Voice.json | 68 canonical command phrases for fuzzy-normalization (`"turn right up"` → `"turn right"`) |
| `stt.garbage_patterns` | config_Voice.json | 17 noise/filler phrases to reject (`"thanks for watching"`, `"okay"`, single letters) |
| `stt.gemini_model` | config_Voice.json | Gemini Live model id (default `gemini-2.5-flash-native-audio-preview-12-2025`); env `MARCUS_GEMINI_MODEL` wins |
| `stt.gemini_api_key` | config_Voice.json | API key fallback (env `MARCUS_GEMINI_API_KEY` or `SANAD_GEMINI_API_KEY` preferred) |
| `stt.gemini_vad_*` | config_Voice.json | server-side VAD start/end sensitivity, prefix padding, silence duration |
| `stt.gemini_session_timeout_sec` | config_Voice.json | reconnect cadence (660 s = Live API session cap) |
| `stt.gemini_record_enabled` | config_Voice.json | save `<ts>_user.wav` per turn under `Data/Voice/Recordings/gemini_turns/` |
| `timeout_ms`, `stale_threshold_s`, `reconnect_delay_s` | config_Camera.json | RealSense frame timeout, reconnect trigger, initial backoff |
| `default_max_steps`, `step_delay_s`, `rotate_speed`, `min_steps_warmup` | config_ImageSearch.json | image-guided search rotation cadence (wired into `Vision/marcus_imgsearch.py`) |
| `default_walk_speed`, `dist_tolerance`, `angle_tolerance`, `safety_timeout_mult`, `dr_update_hz` | config_Odometry.json | precise motion control (wired into `Navigation/marcus_odometry.py`) |
| `MARCUS_LOG_MAX_BYTES`, `MARCUS_LOG_BACKUP_COUNT`, `MARCUS_LOG_DIR` | env vars | log rotation size, backup count, log directory override |

---

## Per-command latency (estimated)

| Step | Typical | Notes |
|---|---|---|
| Mic chunk → Gemini Live | ~32 ms | 512-sample PCM blob over WebSocket |
| Gemini server-side VAD turn-end | ~200 ms | configurable via `gemini_vad_silence_duration_ms` (default 200) |
| Gemini transcript emission | 100–500 ms | depends on utterance length; partials may stream |
| Wake-word check + fuzzy-normalize | <5 ms | `re.search` + difflib against 68 phrases |
| Dispatch dedup | <1 ms | string compare + cooldown |
| Camera frame fetch | <50 ms | poll loop, no 1 s blocking stall |
| Ollama Qwen2.5-VL | 800–1500 ms | `num_batch=128 / num_ctx=2048 / num_predict=120` |
| Executor + ZMQ send | <10 ms | fire-and-forget PUB |
| TtsMaker playback | ~len(text) × 80 ms | synthesizes + plays on robot |
| `flush_mic` × 2 | <1 ms each | bracketed around `audio_api.speak()` |

**Total user-stops-talking → answer-playback:** ~**1.5–3 s** for a short vision question like "Sanad, what do you see" — Gemini's instant turn-detection saves the 2 s "Yes" ack the previous Whisper-era pipeline needed.