BaymaxMini

A State-of-the-Art Healthcare Companion Robot
Real-time biometric monitoring · Expressive face animation · Intelligent medical assessment
Powered by a dual-layer C++17 / Python 3.11 architecture on Raspberry Pi 5
Created by Gabriel Calderon
Requested by Elias Bautista
https://github.com/chele-s/BaymaxMini.git

---

Table of Contents

• Overview
• System Architecture
• Hardware Layer
• Mathematical Foundations
  • Digital Signal Processing
  • Photoplethysmography Pipeline
  • Spring-Damper Dynamics
  • Second-Order Dynamics System
  • Easing & Interpolation Theory
  • Cubic Bézier Curve Solver
  • Power Management Mathematics
  • SpO₂ Computation (Beer-Lambert)
  • INA219 Calibration Mathematics
  • Servo Kinematics
• C++ Real-Time Core
• Python Brain
• IPC Protocol
• Build & Deploy
• Configuration
• Project Structure

---

Overview

BaymaxMini is a healthcare companion robot inspired by the fictional nurse-bot Baymax from Disney's Big Hero 6. It implements a hard real-time control loop in C++17 running at 50 Hz on bare Linux, coupled with a high-level Python brain that handles natural language understanding, computer vision, medical reasoning, and emotional intelligence.

The system runs on a Raspberry Pi 5 and communicates across layers via ZeroMQ PUB/SUB with checksummed binary telemetry frames, ensuring < 1ms IPC latency while maintaining full decoupling between the deterministic hardware layer and the non-deterministic AI layer.

Key Capabilities

Domain	Capability	Implementation
Biometrics	Heart rate, SpO₂, skin temperature	MAX30102 PPG + MLX90614 IR thermometry
Proximity	Time-of-Flight distance sensing	VL53L1X SPAD array (up to 4m)
Expression	10 emotions, blinks, gaze, micro-movements	PCA9685 12-channel servo via spring dynamics
Power	Real-time energy monitoring & protection	INA219 current/voltage sensing with hysteresis FSM
Vision	Person detection, face analysis	YOLOv8n (FP16) + Single-Pass Geometric Deduction
Audio	Speech recognition, TTS, intent parsing	Whisper STT + edge TTS
Medical	Health assessment, medication reminders	Rule-based clinical heuristics + patient history DB
Intelligence	State machine, event bus, scheduling	Finite automaton with priority event dispatch

---

System Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Python Brain (non-RT)                     │
│  ┌─────────┐ ┌──────────┐ ┌────────┐ ┌───────────────────┐ │
│  │ Vision  │ │  Audio   │ │Medical │ │   Core Logic      │ │
│  │ YOLOv8  │ │STT / TTS │ │Assess. │ │ StateMachine      │ │
│  │ FaceAn. │ │IntentPrs │ │History │ │ EventBus/Sched.   │ │
│  └────┬────┘ └────┬─────┘ └───┬────┘ └────────┬──────────┘ │
│       └───────────┴────────────┴───────────────┘            │
│                         │ ZMQ PUB/SUB                       │
│                    tcp://127.0.0.1:5555 (CMD ↓)             │
│                    tcp://127.0.0.1:5556 (TELEM ↑)           │
├─────────────────────────────────────────────────────────────┤
│                  C++ Real-Time Core (50 Hz)                 │
│  ┌─────────────────────────────────────────────────────┐    │
│  │                   Orchestrator                      │    │
│  │   tick() → readSensors → processCmd → logic → out   │    │
│  └──────┬──────────┬──────────┬────────────────────────┘    │
│  ┌──────┴───┐ ┌────┴─────┐ ┌─┴──────────┐                  │
│  │  Face    │ │ Vitals   │ │  Power     │   Modules         │
│  │Controller│ │ Monitor  │ │  System    │                   │
│  └──────┬───┘ └────┬─────┘ └─┬──────────┘                  │
│  ┌──────┴──────────┴─────────┴──────────────────────┐      │
│  │              I²C Bus  (400 kHz)                   │      │
│  │  PCA9685 · VL53L1X · MAX30102 · MLX90614 · INA219│      │
│  └──────────────────────────────────────────────────┘      │
└─────────────────────────────────────────────────────────────┘

---

Hardware Layer

I²C Sensor Suite

Chip	Type	Address	Function	Key Specs
PCA9685	PWM Driver	0x40	12-channel servo control	12-bit resolution, 50 Hz PWM
VL53L1X	ToF Sensor	0x29	Proximity detection	4m range, SPAD array, ±3% accuracy
MAX30102	Pulse Oximeter	0x57	Heart rate + SpO₂	Dual LED (660nm red / 880nm IR), 18-bit ADC
MLX90614	IR Thermometer	0x5A	Contactless body temp	±0.5°C accuracy, 0.02°C resolution
INA219	Power Monitor	0x41	Battery voltage/current	12-bit ADC, ±1% current accuracy

Battery System

• Chemistry: 3S LiPo (Lithium Polymer)
• Nominal: 11.1V · Full: 12.6V · Empty: 9.0V
• Capacity: 2200 mAh
• Protection: Overcurrent 3A trip / 2.5A clear, servo stall 2A / 1.2A, shutdown at 8.5V

---

Mathematical Foundations

1. Digital Signal Processing (DSP)

1.1 Biquad Filter (Second-Order IIR)

The core DSP primitive is a Direct Form I biquad filter, implementing the general second-order transfer function:

$$H(z) = \frac{b_0 + b_1 z^{-1} + b_2 z^{-2}}{1 + a_1 z^{-1} + a_2 z^{-2}}$$

The difference equation computed each sample:

$$y[n] = b_0 \cdot x[n] + b_1 \cdot x[n-1] + b_2 \cdot x[n-2] - a_1 \cdot y[n-1] - a_2 \cdot y[n-2]$$

Coefficients are derived from the analog prototype via the bilinear transform. For a lowpass Butterworth with cutoff frequency $f_c$ and sample rate $f_s$:

$$\omega_0 = 2\pi \frac{f_c}{f_s}, \quad \alpha = \frac{\sin(\omega_0)}{2Q}$$

$$b_0 = \frac{1 - \cos(\omega_0)}{2}, \quad b_1 = 1 - \cos(\omega_0), \quad b_2 = b_0$$

$$a_0 = 1 + \alpha, \quad a_1 = -2\cos(\omega_0), \quad a_2 = 1 - \alpha$$

All coefficients are normalized by $a_0$ before storage. The system supports lowpass, highpass, bandpass, notch, and allpass topologies.

1.2 Cascaded Butterworth (Nth-Order)

For steeper roll-off, biquad sections are cascaded to form higher-order Butterworth filters. An $N$th-order filter uses $N/2$ biquad sections, each with a unique $Q$ factor derived from Butterworth pole placement:

$$\theta_k = \frac{\pi(2k + 1)}{2N}, \quad Q_k = \frac{1}{2\cos(\theta_k)}, \quad k = 0, 1, \ldots, \frac{N}{2} - 1$$

This ensures maximally flat magnitude response in the passband — zero ripple by construction.

1.3 Exponential Moving Average (EMA)

Used throughout for smoothing sensor data with minimal latency:

$$y[n] = \alpha \cdot x[n] + (1 - \alpha) \cdot y[n-1]$$

The smoothing constant $\alpha$ can be derived from a desired cutoff frequency:

$$\alpha = \frac{\Delta t}{RC + \Delta t}, \quad \text{where } RC = \frac{1}{2\pi f_c}$$

1.4 DC Blocker (First-Order Highpass)

Removes DC offset from AC-coupled signals (critical for PPG processing):

$$y[n] = x[n] - x[n-1] + R \cdot y[n-1], \quad R \approx 0.995$$

The pole at $z = R$ creates a highpass with cutoff at:

$$f_c = \frac{f_s}{2\pi} \cdot \arccos\left(\frac{2R}{1 + R^2}\right)$$

1.5 Median Filter (Order 3)

Non-linear filter for impulse noise rejection. For three samples ${a, b, c}$:

$$y = \text{median}(a, b, c)$$

Implemented via a sorting network (3 conditional swaps) for branchless execution on ARM.

1.6 Slew Rate Limiter

Constrains the rate of change to protect servo mechanisms:

$$y[n] = y[n-1] + \text{clamp}\big(x[n] - y[n-1],\ -R_{\text{fall}} \cdot \Delta t,\ R_{\text{rise}} \cdot \Delta t\big)$$

1.7 Absolute Floating-Point Precision (Boost Math)

To eliminate catastrophic cancellation and rounding drift in continuous real-time execution, vital trigonometry and constants strictly leverage boost::math. Phase calculations use exact boost::math::sin_pi() evaluating at the absolute bit-level limit of the hardware architecture.

1.8 O(1) Rolling Variance (Boost Accumulators)

Real-time PPG signal quality tracking utilizes boost::accumulators::accumulator_set, extracting the rolling mean and variance of human vitals in a strictly $O(1)$ time complexity window to guarantee robust noise rejection without manual CPU array-traversal memory overhead.

---

2. Photoplethysmography (PPG) Pipeline

The MAX30102 outputs raw photon count data from red (660nm) and infrared (880nm) LEDs. The full signal processing chain:

Raw IR/Red → Median₃ → DC Removal → Butterworth LPF (5 Hz) → Peak Detection → BPM

2.1 Signal Quality Metric

A composite quality score determines reading validity:

$$Q = 0.6 \cdot Q_{\text{ratio}} + 0.4 \cdot Q_{\text{stability}}$$

Where:

$$Q_{\text{ratio}} = \text{clamp}\left(\frac{\text{peak-to-peak}(AC)}{DC} \cdot 50, \ 0, \ 1\right)$$

$$Q_{\text{stability}} = \text{clamp}\left(1 - \frac{\sigma}{\mu} \cdot 0.5, \ 0, \ 1\right)$$

The ratio score validates adequate perfusion (AC/DC > 0.02), while the stability score penalizes high coefficient of variation.

2.2 Adaptive Peak Detection

Heart beats are detected via zero-crossing of the signal slope with a refractory period and adaptive threshold:

1. Refractory guard: Ignore peaks within 300ms of the last detection
2. Slope analysis: detect sign change $\text{slope}_{n-1} &gt; 0 \land \text{slope}_n \leq 0$
3. Adaptive threshold: $\tau[n] = 0.7 \cdot \tau[n-1] + 0.3 \cdot \text{peak amplitude}$
4. Threshold decay: $\tau[n] \leftarrow 0.998 \cdot \tau[n]$ (adapts to signal loss)
5. BPM computation: $\text{BPM} = \frac{60 \cdot f_s}{\Delta_{\text{samples}}}$, smoothed with EMA ($\alpha = 0.3$)

---

3. Spring-Damper Dynamics

All servo channels (eyelids, gaze, brows, cheeks) are driven by critically damped spring dynamics, ensuring smooth, natural motion without overshoot:

$$\ddot{x} + 2\omega\dot{x} + \omega^2 x = 0$$

The exact closed-form solution for each timestep $\Delta t$:

$$e = \exp(-\omega \Delta t)$$ $$x_{\text{new}} = \big[x + (\dot{x} + \omega x)\Delta t\big] \cdot e$$ $$\dot{x}_{\text{new}} = \big[\dot{x} - \omega(\dot{x} + \omega x)\Delta t\big] \cdot e$$

This provides:

• Zero overshoot (critically damped: $\zeta = 1$)
• Exponential convergence at rate $\omega$ (default $\omega = 12$ rad/s)
• Exact integration — no numerical drift regardless of timestep

The SpringAnimator further extends this with configurable damping ratios for underdamped ($\zeta &lt; 1$, bouncy), critically damped ($\zeta = 1$, smooth), and overdamped ($\zeta &gt; 1$, sluggish) responses. The spring force is integrated via Velocity Verlet:

$$\vec{F} = -k(\vec{x} - \vec{x}_{\text{target}}) - c\dot{\vec{x}}$$

$$\vec{v}_{1/2} = \vec{v}_n + \frac{\vec{a}_n \Delta t}{2}, \quad \vec{x}_{n+1} = \vec{x}_n + \vec{v}_{1/2}\Delta t$$

---

4. Second-Order Dynamics System

A procedural animation system providing frequency-controlled, stability-guaranteed smoothing with anticipation:

$$\ddot{y} = \frac{x + k_3 \dot{x} - y - k_1 \dot{y}}{k_2}$$

Where the constants derive from desired natural frequency $f$, damping $\zeta$, and response $r$:

$$\omega = 2\pi f, \quad k_1 = \frac{\zeta}{\pi f}, \quad k_2 = \frac{1}{\omega^2}, \quad k_3 = \frac{r \cdot \zeta}{2\pi f}$$

The $k_2$ parameter is clamped for unconditional stability:

$$k_2^* = \max\left(k_2, \quad \frac{\Delta t^2}{2} + \frac{\Delta t \cdot k_1}{2}, \quad \Delta t \cdot k_1\right)$$

---

5. Easing & Interpolation Theory

5.1 Hermite Spline

Cubic Hermite interpolation between two points $p_0, p_1$ with tangents $m_0, m_1$:

$$H(t) = h_{00}(t) \cdot p_0 + h_{10}(t) \cdot m_0 + h_{01}(t) \cdot p_1 + h_{11}(t) \cdot m_1$$

With basis functions:

$$h_{00} = 2t^3 - 3t^2 + 1, \quad h_{10} = t^3 - 2t^2 + t$$ $$h_{01} = -2t^3 + 3t^2, \quad h_{11} = t^3 - t^2$$

5.2 Catmull-Rom Spline

A special case of Hermite interpolation with auto-computed tangents:

$$m_i = \frac{p_{i+1} - p_{i-1}}{2}$$

5.3 Smoothstep & Smootherstep

Hermite-basis polynomial interpolation with clamped input:

$$\text{smoothstep}(t) = 3t^2 - 2t^3 \quad \text{(}C^1\text{ continuous)}$$

$$\text{smootherstep}(t) = 6t^5 - 15t^4 + 10t^3 \quad \text{(}C^2\text{ continuous)}$$

5.4 Complete Easing Library (32 functions)

Family	Mathematical Form
Sine	$f(t) = 1 - \cos\left(\frac{\pi t}{2}\right)$
Quadratic	$f(t) = t^2$
Cubic	$f(t) = t^3$
Quartic	$f(t) = t^4$
Quintic	$f(t) = t^5$
Exponential	$f(t) = 2^{10(t-1)}$
Circular	$f(t) = 1 - \sqrt{1 - t^2}$
Back	$f(t) = t^2[(s+1)t - s], \quad s = 1.70158$
Elastic	$f(t) = -2^{10t-10}\sin\left(\frac{(10t - 10.75) \cdot 2\pi}{3}\right)$
Bounce	Piecewise parabolic with $n_1 = 7.5625$

Each family provides in, out, and inOut variants (30 total), plus smoothstep and smootherstep.

---

6. Cubic Bézier Curve Solver

CSS-style timing functions via parametric cubic Bézier $B(t) = (X(t), Y(t))$:

$$X(t) = 3(1-3x_2+3x_1)t^3 + 3(3x_2-6x_1)t^2 + 3x_1 t$$

Given a target $x$, the parameter $t$ is found via Newton-Raphson iteration (8 steps) with bisection fallback (20 steps):

$$t_{n+1} = t_n - \frac{X(t_n) - x}{X'(t_n)}, \quad \text{tolerance} = 10^{-7}$$

Pre-built presets: easeInOut, snappy, organic, overshoot, rubberBand, anticipate, heavyLanding.

---

7. Power Management Mathematics

7.1 Battery State-of-Charge (SoC)

Estimated via a piecewise-linear lookup table against cell voltage, derived from empirical Li-ion discharge curves:

Cell V	3.00	3.30	3.50	3.60	3.70	3.75	3.80	3.85	3.95	4.10	4.20
SoC %	0	5	10	20	30	40	50	60	70	90	100

Cell voltage derived from bus voltage: $V_{\text{cell}} = V_{\text{bus}} / N_{\text{cells}}$

7.2 Runtime Estimation

$$t_{\text{remaining}} = \frac{C_{\text{mAh}} \cdot \text{SoC} / 100}{\bar{I}_{\text{mA}}}$$

7.3 Hysteresis State Machine

All protection thresholds use Schmitt-trigger hysteresis to prevent chattering:

        ┌─────────────────┐
        │   INACTIVE       │
        │                  │ value > trip_threshold
        │    ────────────► │──────────────────────►┐
        │                  │                        │
        └─────────────────┘                        ▼
                  ▲                         ┌──────┴─────┐
                  │  value < clear_threshold│   ACTIVE    │
                  └─────────────────────────│             │
                                            └─────────────┘

---

8. SpO₂ Computation (Beer-Lambert Law)

Blood oxygen saturation is derived from the ratio of ratios (R) of the red and infrared PPG signals:

$$R = \frac{AC_{\text{red}} / DC_{\text{red}}}{AC_{\text{IR}} / DC_{\text{IR}}}$$

SpO₂ is then estimated via the empirically calibrated linear approximation:

$$\text{SpO}_2 = 110 - 25R$$

This derives from the Beer-Lambert law of optical absorption:

$$I = I_0 \cdot e^{-\epsilon(\lambda) \cdot c \cdot d}$$

Where $\epsilon(\lambda)$ is the molar extinction coefficient at wavelength $\lambda$, $c$ is concentration, and $d$ is optical path length. The dual-wavelength ratiometric approach cancels out path length and tissue geometry.

---

9. INA219 Calibration Mathematics

The INA219 current sense amplifier requires precise register calibration:

$$\text{Current_LSB} = \frac{I_{\text{max_expected}}}{2^{15}} \quad [\text{A/bit}]$$

$$\text{CAL} = \text{trunc}\left(\frac{0.04096}{I_{\text{LSB}} \cdot R_{\text{shunt}}}\right)$$

$$\text{Power_LSB} = 20 \times \text{Current_LSB} \quad [\text{W/bit}]$$

With the configured $R_{\text{shunt}} = 0.1\Omega$ and $I_{\text{max}} = 3.2$A:

$$I_{\text{LSB}} = \frac{3.2}{32768} \approx 97.66 ; \mu\text{A/bit}$$

---

10. Servo Kinematics

Pulse Width Mapping

Servo angle $\theta$ maps to pulse width via linear interpolation:

$$\text{PW}(\theta) = \text{PW}_{\min} + \frac{\theta - \theta_{\min}}{\theta_{\max} - \theta_{\min}} \cdot (\text{PW}_{\max} - \text{PW}_{\min})$$

Default range: $500\mu s$ (0°) → $2500\mu s$ (180°)

PCA9685 Tick Conversion

$$\text{ticks} = \text{round}\left(\frac{\text{PW}_{\mu s} \cdot f_{\text{PWM}} \cdot 4096}{10^6}\right)$$

At $f_{\text{PWM}} = 50$Hz: $\text{ticks} = \text{PW}_{\mu s} \times 0.2048$

---

C++ Real-Time Core (src/cpp_core)

Tick Architecture

The core runs a fixed-timestep loop at 50 Hz ($\Delta t = 20$ms) with overrun detection:

while (running) {
    nextTick += 20ms
    readSensors()        ─► I²C reads from all 5 chips
    processCommands()    ─► ZMQ SUB (non-blocking)
    updateLogic()        ─► FSM + spring dynamics + animation
    writeOutputs()       ─► Telemetry PUB + servo PWM
    sleep_until(nextTick)
    if (overrun > 3 ticks) { resync; overrun_count++ }
}

State Machine

    BOOT ──(ZMQ connected)──► CONNECTED ──(brain timeout 2s)──► AUTONOMOUS
      ▲                           ▲                                  │
      │                           └──────(CMD received)──────────────┘
      │
      └──────────────────── SHUTDOWN ◄──(SIGINT/SIGTERM or critical battery)

Module: FaceController

Manages 12 servo channels mapped to facial features via spring-damped dynamics:

Channel	Servo	Feature	Spring ω
0-1	Left eyelids	Upper/Lower lid	12 rad/s
2-3	Right eyelids	Upper/Lower lid	12 rad/s
4-5	Left gaze	Horizontal/Vertical	12 rad/s
6-7	Right gaze	Horizontal/Vertical	12 rad/s
8-9	Cheeks	Left/Right raise	12 rad/s
10-11	Brows	Left/Right position	12 rad/s

10 Expression Poses: Neutral, Happy, Sad, Surprised, Angry, Sleepy, Concerned, Curious, Love, Thinking — each defined as a target pose for all 12 channels with smooth spring-interpolated transitions.

Blink System: Asymmetric timing modeled after human physiology — close: 75ms, hold: 35ms, open: 130ms, ratio 1.73:1 (close:open speed). Auto-blinks at 2-6s intervals with 18% variance.

Micro-Movements: Continuous subtle animation to avoid the uncanny valley:

• Gaze jitter: 0.02 amplitude, 0.3 Hz
• Gaze drift: 0.05 amplitude, 0.08 Hz
• Lid twitch: 0.015 amplitude, 0.15 Hz
• Brow drift: 0.01 amplitude, 0.06 Hz

Module: VitalsMonitor

State machine for clinical-grade vital sign acquisition:

IDLE → DETECTING_FINGER → STABILIZING (2s) → MEASURING (5-10s) → COMPLETE
                                                    ↓
                                         ERROR_NO_FINGER / ERROR_POOR_SIGNAL

Health assessment heuristics:

Condition	Threshold	Assessment
Normal HR	50-110 BPM	NORMAL
Tachycardia	> 110 BPM	ELEVATED_HR
Bradycardia	< 50 BPM	LOW_HR
Hypoxemia	SpO₂ < 95%	LOW_SPO2
Fever	> 38.0°C	FEVER
Hypothermia	< 36.0°C	HYPOTHERMIA
Critical	HR < 40 or > 180, SpO₂ < 90%	CRITICAL

Module: PowerSystem

Real-time energy management with multi-layered protection:

• EMA-filtered voltage, current, power readings
• Coulomb counting for energy accumulation: $E += P \cdot \Delta t$
• 7-state FSM: NOMINAL → LOW_BATTERY → CRITICAL → OVERCURRENT → SERVO_STALL → OVERTEMP → SHUTDOWN
• Callbacks for face alert mode and emergency shutdown

---

Python Brain (src/python_brain)

Subsystem	Modules	Purpose
core	brain.py, event_bus.py, logger.py	Central coordinator, pub/sub events, structured logging
vision	camera.py, detector.py, face_analyzer.py, object_detector.py	Single YOLOv8n FP16 orchestrator, CPU-light geometric facial deduction
audio	stt_engine.py, tts_engine.py, intent_parser.py, sound_fx.py, audio_alerts.py	Speech-to-text, text-to-speech, NLU intent extraction, sound effects
medical	patient_history.py, pharmacist.py, scheduler.py	Patient records, medication database, reminder scheduling
communication	zmq_link.py, telemetry.py	ZMQ bridge to C++ core, telemetry deserialization
logic	state_machine.py, scheduler.py	Behavioral FSM, task scheduling
config	settings.py, db_migrate.py	YAML config loader, database migrations
utils	time_utils.py	Timezone-aware time helpers

---

IPC Protocol

Telemetry Frame (C++ → Python, 50 Hz)

Binary packed struct with djb2 checksum verification:

Field	Type	Offset	Description
magic	u32	0	0xBABE0001
version	u32	4	Protocol version
timestamp_us	u64	8	Microsecond timestamp
sequence	u32	16	Monotonic counter
distance_mm	f32	20	VL53L1X range
heart_rate_bpm	f32	24	Filtered heart rate
spo2_percent	f32	28	Blood oxygen %
skin_temp_c	f32	32	Body temperature
bus_voltage_v	f32	40	Battery voltage
battery_pct	f32	52	State of charge
eyelid_openness	f32	56	Current lid state
gaze_x / gaze_y	f32	60, 64	Eye direction
state	u8	—	System state enum
expression	u8	—	Current face expression
alert	u8	—	Alert level
checksum	u32	last	djb2 integrity check

Command Frame (Python → C++)

Supports: SET_EXPRESSION, SET_EYELID, TRIGGER_BLINK, SET_GAZE, SET_BREATH, SPEAK, SET_LED_COLOR, PLAY_ANIMATION, EMERGENCY_STOP, SHUTDOWN

---

Build & Deploy

Prerequisites

sudo apt-get install build-essential cmake libzmq3-dev nlohmann-json3-dev
pip install pyzmq pyyaml numpy opencv-python-headless tflite-runtime

Build C++ Core

mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j$(nproc)

Run

./build/baymax_core &
python3 src/python_brain/main.py

---

Configuration

File	Purpose
configs/sensors_config.yaml	I²C addresses, sample rates, filter parameters
configs/vision_params.yaml	YOLO confidence thresholds, camera resolution
configs/reminders.json	Medication and appointment schedule

---

Project Structure

BaymaxMini/
├── CMakeLists.txt
├── README.md
├── requirements.txt
├── configs/
│   ├── sensors_config.yaml
│   ├── vision_params.yaml
│   └── reminders.json
├── models/
│   └── yolo_nano_int8.tflite
├── scripts/
├── src/
│   ├── cpp_core/
│   │   ├── main.cpp                    Orchestrator + tick loop
│   │   ├── drivers/
│   │   │   ├── I2C_Bus.{cpp,h}        Linux I²C ioctl wrapper
│   │   │   ├── PCA9685.{cpp,h}        16-ch PWM servo driver
│   │   │   ├── VL53L1X.{cpp,h}        Time-of-Flight ranging
│   │   │   ├── MAX30102.{cpp,h}       Pulse oximetry + HR
│   │   │   ├── MLX90614.{cpp,h}       IR thermometry
│   │   │   └── INA219.{cpp,h}         Power monitoring
│   │   ├── modules/
│   │   │   ├── FaceController.{cpp,h}  Expressive servo animation
│   │   │   ├── VitalsMonitor.{cpp,h}   Clinical vitals pipeline
│   │   │   └── PowerSystem.{cpp,h}     Energy management FSM
│   │   ├── interfaces/
│   │   │   └── I_Sensor.h              Abstract sensor + registry
│   │   ├── ipc/
│   │   │   └── SharedData.h            Binary IPC protocol
│   │   └── utils/
│   │       ├── MathUtils.h             DSP, filters, springs, Vec2
│   │       ├── DigitalFilter.h         PPG chain, peak detection
│   │       └── Easing.h                Animation, Bézier, timelines
│   └── python_brain/
│       ├── main.py                     Brain entry point
│       ├── core/                       Event bus, state, logging
│       ├── vision/                     Camera, YOLO, face analysis
│       ├── audio/                      STT, TTS, intent parsing
│       ├── medical/                    Patient history, meds
│       ├── communication/             ZMQ link, telemetry decode
│       ├── logic/                      Behavioral state machine
│       ├── config/                     Settings, DB migration
│       └── utils/                      Time utilities

---

"Hello. I am Baymax, your personal healthcare companion."