Buff Flow — Motion Capture Setup Guide

Buff Flow — Motion Capture Setup Guide

Document ID: BUFF-FLOW-MOCAP-001
Version: 1.0.0
Last Updated: 2026-01-18

---

Overview

This guide covers setting up Sony Mocopi motion capture for Buff Flow sessions. Motion data enables gesture recognition, form feedback, and the computational choreography features in later phases.

---

Equipment

Required

Recommended

---

Sony Mocopi Overview

What It Is

Sony Mocopi is a 6-sensor IMU (Inertial Measurement Unit) system that captures full-body motion without cameras. Each sensor tracks rotation and acceleration, and the system reconstructs a 27-bone skeleton.

Sensors

Specifications

• Frame Rate: 60 Hz
• Latency: ~50ms
• Battery Life: ~10 hours per sensor
• Connectivity: Bluetooth LE
• Output: 27-bone skeleton (quaternion + position)

---

Sensor Placement for Buff Flow

Standard Placement

        ┌───┐
        │ H │  ← Head sensor (headband)
        └───┘
          │
    ┌─────┼─────┐
    │     │     │
   LW    [T]    RW  ← Wrists (watch bands), Torso (chest)
    │           │
    │           │
    │           │
   LA          RA  ← Ankles (straps)

Placement Tips for Buff Flow

Head:
- Use sport headband, not cap (allows sweating)
- Sensor sits on forehead or side
- Tight enough to not shift during jumps

Torso:
- Center of chest, under apron if wearing
- Strap should be snug but allow breathing
- May need to adjust after warm-up

Wrists:
- Inside wrist preferred (less movement artifact)
- Below wristbone, above hand
- Secure when holding weights

Ankles:
- Above ankle bone, below calf
- Tight enough to not slip during jumps
- Check after warm-up

Weight Considerations

When dancing with weights, wrist sensors may shift:
- Use tighter bands
- Consider wrist wraps over sensors
- Check alignment between songs

---

Software Setup

cc-mcs (Motion Capture Server)

Location: `[home]/Desktop/Comp-Core/backend/cc-mcs/`

Starting the Server:

cd [home]/Desktop/Comp-Core/backend/cc-mcs
cargo run --release

Default Ports:
- Mocopi UDP: 12351
- WebSocket: 8765
- HTTP API: 3847

Mocopi App Connection

1. Open Sony Mocopi app on phone
2. Pair all 6 sensors
3. In app settings, enable "External Output"
4. Set destination IP to laptop
5. Set port to 12351 (UDP)

Verifying Connection

# Check if data is flowing
curl http://localhost:3847/status

# Should show:
# {
#   "connected": true,
#   "sensors": 6,
#   "fps": 60,
#   "session_id": "..."
# }

---

Recording Workflow

Pre-Session Checklist

30 Minutes Before:
- [ ] Charge all 6 Mocopi sensors (check in app)
- [ ] Start laptop, open cc-mcs
- [ ] Create session folder: `YYYY-MM-DD_tier_name`
- [ ] Test Bluetooth connection

10 Minutes Before:
- [ ] Put on sensors
- [ ] Calibrate in Mocopi app (T-pose)
- [ ] Verify all 6 showing in cc-mcs
- [ ] Test range of motion

2 Minutes Before:
- [ ] Start recording in cc-mcs
- [ ] Say "Recording started" (audio marker)
- [ ] Begin session

During Session

Recording Controls:
- Start: `curl -X POST http://localhost:3847/record/start`
- Stop: `curl -X POST http://localhost:3847/record/stop`
- Mark: `curl -X POST http://localhost:3847/mark -d '{"label":"The Barista"}'`

Verbal Markers:
Say move names out loud during transitions:
- "Starting The Barista"
- "Into Coffee Grind"
- "Espresso Shot now"

This helps with labeling later (if also recording audio).

Sensor Issues:
If a sensor disconnects:
1. Continue session if possible
2. Note the time
3. Reconnect during recovery break
4. Re-calibrate if needed

Post-Session

Immediately After:
1. Stop recording
2. Verify file saved
3. Quick playback check
4. Note any issues in session log

Export:

# Export to standard format
cc-mcs export --session SESSION_ID --output /path/to/sessions/

# Creates:
# - session.json (metadata)
# - frames.bin (motion data)
# - markers.json (labeled points)

---

Data Organization

Directory Structure

[home]/spine/Buf Barista/data/motion/
├── sessions/
│   ├── 2026-03-15_full-roast_001/
│   │   ├── session.json
│   │   ├── frames.bin
│   │   ├── markers.json
│   │   └── notes.txt
│   ├── 2026-03-17_first-brew_001/
│   └── ...
├── labeled/
│   ├── the-barista/
│   │   ├── sample_001.json
│   │   ├── sample_002.json
│   │   └── ...
│   ├── coffee-grind/
│   └── ...
└── motionphrase/
    └── buff_flow_library.json

Session Naming Convention

YYYY-MM-DD_tier_sequence

Examples:
- `2026-03-15_full-roast_001`
- `2026-03-15_full-roast_002`
- `2026-03-17_first-brew_001`

Session Log Template

Create `notes.txt` for each session:

Session: 2026-03-15_full-roast_001
Date: March 15, 2026
Tier: Full Roast (45 min)
Duration: 47:23

Sensors:
- All 6 connected throughout
- Left wrist strap loosened at 23:00, retightened

Issues:
- Brief dropout at 31:45 (~5 seconds)
- Head sensor shifted during jumps

Moves Performed:
- The Barista: Multiple throughout
- Coffee Grind: 3x in Flow Block 2
- Espresso Shot: 2x, once in each block

Notes:
- Good energy, clear move transitions
- New combo worth labeling: Flow Press → Coffee Grind

---

Labeling Moves

Using cc-mcs Playback

# Start playback tool
cc-mcs playback --session 2026-03-15_full-roast_001

# Opens browser interface at localhost:3847/playback

Labeling Interface

1. Play/pause with spacebar
2. Use frame-by-frame arrows
3. Mark start frame (S key)
4. Mark end frame (E key)
5. Enter move name
6. Save label

Move Label Schema

{
  "move_name": "the_barista",
  "display_name": "The Barista",
  "session_id": "2026-03-15_full-roast_001",
  "start_frame": 3450,
  "end_frame": 3690,
  "duration_ms": 4000,
  "quality": "good",
  "notes": "Clean execution, full range"
}

Quality Ratings

---

Building MotionPhrase Library

What Is a MotionPhrase?

A MotionPhrase is a reusable motor program — a labeled segment of motion that can be:
- Retrieved by similarity (RAG++)
- Used for gesture classification training
- Compared for form feedback

Export to MotionPhrase Format

# From labeled moves
cc-mcs build-phrases \
  --input /path/to/labeled/ \
  --output /path/to/motionphrase/buff_flow_library.json

# Options
--min-quality good       # Minimum quality to include
--normalize true         # Normalize to standard skeleton
--include-anticipation   # Add anticipation signals

Library Structure

{
  "library_name": "buff_flow_v1",
  "created": "2026-06-01",
  "moves": {
    "the_barista": {
      "samples": 15,
      "avg_duration_ms": 4000,
      "embeddings": [...],
      "anticipation_profile": {
        "commitment_curve": [...],
        "transition_points": [...]
      }
    },
    "coffee_grind": {...},
    ...
  }
}

---

Troubleshooting

Sensor Battery Indicators

---

Backup Camera (MediaPipe)

When to Use

• Mocopi sensors unavailable
• Need hand/face data (not in Mocopi)
• Backup for sensor failures
• Testing without full setup

Setup

1. Position iPhone/iPad on tripod
2. Full body visible in frame
3. Good lighting (avoid backlight)
4. Start MediaPipe capture in cc-mcs

cc-mcs --mediapipe --camera-id 0

Limitations vs Mocopi

---

Data Privacy

What Gets Recorded

• Skeleton positions/rotations only
• No video or images (unless MediaPipe backup)
• Session metadata (date, duration, tier)
• Move labels

Storage

• Local only (laptop + external SSD)
• No cloud upload unless explicitly enabled
• Backup to encrypted drive recommended

---

Related Documentation

• [COMP_CORE_INTEGRATION.md](COMP_CORE_INTEGRATION.md) — Full integration plan
• [BUFF_FLOW_PROGRAM.md](BUFF_FLOW_PROGRAM.md) — Program overview
• [CHOREOGRAPHY_LIBRARY.md](CHOREOGRAPHY_LIBRARY.md) — Move definitions
• Comp-Core docs: `/Desktop/Comp-Core/core/motion/cc-collection/docs/`

---

Change Log