Split View: Pixhawk & eVTOL 비행제어 완전 정복: PX4, 칼만 필터, PWM 신호, 센서 퓨전부터 ROS2 통합까지

Pixhawk & eVTOL Flight Control Mastery: From PX4, Kalman Filters, PWM Signals, and Sensor Fusion to ROS2 Integration

• 1. Overview
• 2. Pixhawk — The Standard in Open-Source Flight Controllers
  • 2.1 What Is Pixhawk?
  • 2.2 FMU Version History
  • 2.3 Pixhawk 6X (FMUv6X) Detailed Specifications
  • 2.4 Pixhawk 6C (FMUv6C)
  • 2.5 Supported Autopilot Firmware
  • 2.6 Ecosystem
• 3. Flight Control System Architecture
  • 3.1 Cascaded Control Structure
  • 3.2 PID Controller
  • 3.3 Position/Velocity Controller
  • 3.4 Attitude/Rate Controller
  • 3.5 Mixer (Control Allocation)
• 4. PWM Signals and ESC Protocols
  • 4.1 PWM (Pulse Width Modulation) Fundamentals
  • 4.2 ESC Protocol Comparison
  • 4.3 DShot Protocol in Detail
  • 4.4 PWM vs DShot — Key Differences
• 5. Kalman Filter — The Core Algorithm of Flight Control
  • 5.1 Kalman Filter Fundamentals
  • 5.2 Extended Kalman Filter (EKF)
  • 5.3 Error-State Kalman Filter (ESKF)
  • 5.4 PX4 EKF2 (ECL EKF) — 24-State Vector
  • 5.5 ArduPilot EKF2 vs EKF3
  • 5.6 Complementary Filter vs Kalman Filter
• 6. eVTOL (Electric Vertical Takeoff and Landing)
  • 6.1 What Is eVTOL?
  • 6.2 Market Overview
  • 6.3 Airframe Configuration Types
    • Multirotor
    • Lift + Cruise
    • Tiltrotor
    • Tiltwing
  • 6.4 Certification Framework
  • 6.5 Battery Technology
• 7. PX4 Autopilot
  • 7.1 Architecture
  • 7.2 Flight Modes
  • 7.3 MAVLink Protocol
• 8. ArduPilot
  • 8.1 PX4 vs ArduPilot
  • 8.2 Supported Vehicles
• 9. ROS2 Integration
  • 9.1 Two Integration Approaches
  • 9.2 Offboard Control Example
• 10. SITL/HITL Simulation
  • 10.1 SITL (Software-In-The-Loop)
  • 10.2 HITL (Hardware-In-The-Loop)
  • 10.3 Gazebo Integration
• 11. Learning Roadmap
  • 11.1 Beginner (1-2 Months)
  • 11.2 Intermediate (2-4 Months)
  • 11.3 Advanced (4-6 Months)
• 12. References
  • Official Documentation
  • Hardware
  • Control and Estimation
  • ESC Protocols
  • eVTOL
  • ROS2 Integration
  • GitHub Repositories
• Quiz

1. Overview

Drones and eVTOLs (electric Vertical Takeoff and Landing aircraft) are no longer experimental technology. In 2026, Joby Aviation and Archer Aviation are launching commercial operations, and Urban Air Mobility (UAM) is becoming a reality. At the heart of all this technology lies the flight control system.

This post provides a comprehensive summary of the core elements of flight control.

┌────────────────────────────────────────────────────────────┐
│            Flight Control System Architecture               │
│                                                             │
│  Sensor Layer     Estimation Layer   Control Layer  Output  │
│  ┌─────────┐    ┌──────────┐    ┌──────────┐    ┌────────┐│
│  │IMU(6DoF)│    │  Kalman   │    │PID Control│    │  PWM/  ││
│  │GPS      │──→│ Filter    │──→│ Pos→Vel   │──→│ DShot  ││
│  │Baro     │    │(EKF2/3)  │    │ →Att→Rate │    │  ESC   ││
│  │Mag      │    │Sen. Fusion│    │ →Mixer   │    │ Motors ││
│  └─────────┘    └──────────┘    └──────────┘    └────────┘│
│                                                             │
│  Comms: MAVLink ←→ QGroundControl / ROS2 / Companion PC    │
└────────────────────────────────────────────────────────────┘

2. Pixhawk — The Standard in Open-Source Flight Controllers

2.1 What Is Pixhawk?

Pixhawk is not a specific product but an open hardware standard. It is maintained by the Dronecode Foundation (under the Linux Foundation), and over one million Pixhawk-based devices are in operation worldwide.

2.2 FMU Version History

2.3 Pixhawk 6X (FMUv6X) Detailed Specifications

Processors:

• FMU: STM32H753 (Cortex-M7, 480MHz, 2MB Flash, 1MB RAM)
• IO: STM32F103 (Cortex-M3, 72MHz, 64KB SRAM)

Onboard Sensors (Triple Redundant):

Key Interfaces:

• 16+ PWM servo outputs (8 IO + 8 FMU)
• Ethernet PHY — high-speed communication with mission computers
• 4x UART, 3x SPI, 2x CAN, 2x I2C
• USB-C, RC input (SBUS/CPPM/DSM)

Modular Design: IMU Board + FMU Board + Base Board separated via 100-pin/50-pin Pixhawk Autopilot Bus connectors

2.4 Pixhawk 6C (FMUv6C)

2.5 Supported Autopilot Firmware

2.6 Ecosystem

Manufacturers: Holybro, CUAV, ARK Electronics, ModalAI, NXP, mRo, Auterion Tools: QGroundControl (GCS), MAVLink, MAVSDK, MAVROS, ROS2 integration

3. Flight Control System Architecture

3.1 Cascaded Control Structure

The PX4 multicopter controller uses a cascaded position-velocity-attitude-rate loop.

[Pos SP] → [Pos P] → [Vel PID] → [Att P] → [Rate K-PID] → [Mixer] → [Motors]
  Outer loop           Mid loop      Att loop     Inner loop
  ~50Hz               ~50Hz         ~250Hz        ~1000Hz

3.2 PID Controller

Continuous-Time PID:

$u(t) = K_p \cdot e(t) + K_i \int_0^t e(\tau) d\tau + K_d \frac{de(t)}{dt}$

Discrete-Time PID (flight controller implementation):

$u[k] = K_p \cdot e[k] + K_i \sum_{j=0}^{k} e[j] \cdot \Delta t + K_d \frac{e[k] - e[k-1]}{\Delta t}$

3.3 Position/Velocity Controller

Position Controller (P):

$\vec{v}_{sp} = K_p^{pos} \cdot (\vec{p}_{sp} - \vec{p}_{est})$

• PX4 parameters: MPC_XY_P, MPC_Z_P
• Velocity saturation: MPC_XY_VEL_MAX

Velocity Controller (PID): Velocity error to acceleration setpoint, with anti-reset windup applied

3.4 Attitude/Rate Controller

Attitude Controller (P, quaternion-based):

$\vec{\omega}_{sp} = K_p^{att} \cdot \vec{e}_{att}$

Rate Controller (K-PID): Low-pass filter applied to the derivative path for noise reduction

Gyroscope Data Processing Pipeline:

1. Calibration correction -> 2. Bias removal -> 3. Notch filter -> 4. Low-pass filter

3.5 Mixer (Control Allocation)

Converts rate controller outputs into individual motor commands.

Quadcopter X-configuration example:

$\begin{bmatrix} M_1 \\ M_2 \\ M_3 \\ M_4 \end{bmatrix} = \begin{bmatrix} 1 & -1 & 1 & 1 \\ 1 & 1 & -1 & 1 \\ 1 & 1 & 1 & -1 \\ 1 & -1 & -1 & -1 \end{bmatrix} \begin{bmatrix} T \\ \tau_x \\ \tau_y \\ \tau_z \end{bmatrix}$

Where $T$ = total thrust, $\tau_x$ = roll torque, $\tau_y$ = pitch torque, $\tau_z$ = yaw torque

4. PWM Signals and ESC Protocols

4.1 PWM (Pulse Width Modulation) Fundamentals

PWM is a technique that conveys analog values by modulating the duty cycle of a digital signal.

     ┌────┐              ┌────────┐
     │    │              │        │
─────┘    └──────────────┘        └──────────
     |<-->|              |<------>|
     1000us              2000us
     (0% throttle)       (100% throttle)
     |<-------------------->|
         20ms (50Hz period)

4.2 ESC Protocol Comparison

4.3 DShot Protocol in Detail

DShot is the de facto standard ESC protocol as of 2025.

Frame Structure (16 bits):

[11-bit throttle (0-2047)] [1-bit telemetry request] [4-bit CRC]

• Throttle values 1-47: Reserved for special commands (beep, motor direction reversal, ESC settings)
• Bit encoding: 0/1 distinguished by HIGH/LOW ratio (1 = 75% HIGH, 0 = 37.5% HIGH)

Bidirectional DShot:

• ESC sends eRPM telemetry back to the FC
• Real-time RPM data enables dynamic notch filtering, dramatically improving vibration suppression
• Data provided: RPM, voltage, current, temperature

4.4 PWM vs DShot — Key Differences

5. Kalman Filter — The Core Algorithm of Flight Control

5.1 Kalman Filter Fundamentals

The Kalman filter is a recursive algorithm that optimally estimates system state from noisy multi-sensor measurements.

┌──────────────────────────────────────────────────────┐
│              Kalman Filter Cycle                       │
│                                                       │
│  ┌──────────────────┐     ┌──────────────────┐       │
│  │  Step 1: Predict  │     │  Step 2: Update   │       │
│  │  (Time Update)    │────►│  (Meas. Update)  │       │
│  │                  │     │                  │       │
│  │  x̂ = F·x + B·u  │     │  K = P·Hᵀ·S⁻¹   │       │
│  │  P = F·P·Fᵀ + Q │     │  x̂ = x̂ + K·(z-Hx̂)│       │
│  │                  │     │  P = (I-KH)·P    │       │
│  │ Predict via      │     │ Correct with     │       │
│  │ motion model     │     │ measurements     │       │
│  └──────────────────┘     └────────┬─────────┘       │
│           ▲                        │                  │
│           └────────────────────────┘                  │
│                    Repeat                             │
└──────────────────────────────────────────────────────┘

Step 1: Prediction (Time Update)

$\hat{x}_{k|k-1} = F_k \hat{x}_{k-1|k-1} + B_k u_k$

$P_{k|k-1} = F_k P_{k-1|k-1} F_k^T + Q_k$

Step 2: Update (Measurement Update)

$K_k = P_{k|k-1} H_k^T (H_k P_{k|k-1} H_k^T + R_k)^{-1}$

$\hat{x}_{k|k} = \hat{x}_{k|k-1} + K_k (z_k - H_k \hat{x}_{k|k-1})$

$P_{k|k} = (I - K_k H_k) P_{k|k-1}$

Intuition Behind the Kalman Gain:

• $K \to 1$: Trust the measurement more (low measurement noise)
• $K \to 0$: Trust the prediction more (accurate model)

5.2 Extended Kalman Filter (EKF)

Since drone dynamics are nonlinear, the standard Kalman filter cannot be applied directly. The EKF linearizes nonlinear functions using a first-order Taylor series expansion.

Nonlinear System:

$x_k = f(x_{k-1}, u_k) + w_k \quad \text{(state transition)}$

$z_k = h(x_k) + v_k \quad \text{(observation)}$

EKF Prediction:

$\hat{x}_{k|k-1} = f(\hat{x}_{k-1|k-1}, u_k)$

$P_{k|k-1} = F_k P_{k-1|k-1} F_k^T + Q_k, \quad F_k = \frac{\partial f}{\partial x}\bigg|_{\hat{x}_{k-1}}$

EKF Update:

$K_k = P_{k|k-1} H_k^T (H_k P_{k|k-1} H_k^T + R_k)^{-1}, \quad H_k = \frac{\partial h}{\partial x}\bigg|_{\hat{x}_{k|k-1}}$

5.3 Error-State Kalman Filter (ESKF)

PX4's EKF2 implements the Error-State Kalman Filter. It is essential for estimating uncertainty in quaternion rotations.

Key Idea:

• Instead of directly estimating the quaternion, estimate the error rotation vector
• Represent rotational uncertainty as a 3D vector in the tangent space of SO(3)
• Separate the Nominal State (inertial navigation equations) from the Error State (Kalman filter)

$\delta x = x_{true} - \hat{x}_{nominal}$

5.4 PX4 EKF2 (ECL EKF) — 24-State Vector

Sensor Fusion Architecture:

• Delayed fusion: Sensor data stored in a FIFO buffer to accommodate per-sensor time delays, fusing data at matching timestamps
• IMU data: Used only for state prediction, never as an observation
• Multiple EKF instances: Enhanced fault detection (MAX(IMU count, 1) x MAX(magnetometer count, 1), up to 16 instances)
• Multi-hypothesis filter: Gaussian sum filter for yaw estimation, capable of using GPS velocity alone

5.5 ArduPilot EKF2 vs EKF3

5.6 Complementary Filter vs Kalman Filter

$\theta_{est} = \alpha \cdot (\theta_{est} + \omega_{gyro} \cdot \Delta t) + (1-\alpha) \cdot \theta_{accel}$

Where $\alpha = \frac{\tau}{\tau + \Delta t}$ (typically 0.96-0.98)

6. eVTOL (Electric Vertical Takeoff and Landing)

6.1 What Is eVTOL?

An eVTOL is an aircraft that takes off and lands vertically using an electric propulsion system and flies horizontally. As a core element of Urban Air Mobility (UAM), it is ushering in the era of "flying taxis."

6.2 Market Overview

• Market size: ~$4.2B in 2025 to ~$8.0B by 2033 (CAGR 37%)
• Commercial operations expected to begin in 2026: Joby Aviation, Archer Aviation

6.3 Airframe Configuration Types

Multirotor

    ①        ②
     \      /
      ┌────┐
      │    │
      └────┘
     /      \
    ③        ④

• 4-8 rotors provide both lift and thrust
• Pros: Simple, stable hovering
• Cons: Worst energy efficiency (inefficient during cruise)
• Examples: EHang EH216-S, Volocopter VoloCity

Lift + Cruise

    Vertical rotor  Vertical rotor        Rear propeller
         ↓              ↓                     ↓
       ┌──┐          ┌──┐    ═══════    ▶───
       └──┘          └──┘      Wing

• Separate rotors for vertical takeoff and propellers for horizontal flight
• Pros: Each flight phase is optimized
• Cons: Vertical rotors are dead weight during cruise
• Examples: Archer Midnight, Vertical Aerospace VX4

Tiltrotor

  Hover:   →  Cruise:
  ↑  ↑       →  →
  │  │       ╱  ╲
  ┌──┐      ┌──────┐
  └──┘      └──────┘

• Rotors tilt between vertical and horizontal orientations
• Pros: Efficient with a single propulsion system
• Cons: Mechanical complexity of tilt mechanism
• Examples: Joby S4 (6 tilting rotors), Bell Nexus

Tiltwing

• The wing itself tilts along with the rotors
• High-speed, long-range capability
• Examples: Lilium Jet (36 electric engines embedded in wings)

6.4 Certification Framework

FAA (United States):

• eVTOL reclassified under the "Powered-Lift" category
• AC 21.17-4 (2025.07): Max 12,500 lbs, 6 seats or fewer, battery-electric propulsion
• Joby: Stage 4 certification 70% complete (2025 Q3)
• Archer: Achieved longest piloted flight of 55 miles (2025.08)

EASA (Europe):

• Issued SC-VTOL (Special Condition - VTOL)
• Pursuing standards harmonization with FAA

6.5 Battery Technology

Solid-State Battery Breakthrough:

• EHang + Inx: Achieved 480 Wh/kg
• World's first eVTOL solid-state battery flight test successful (48 min 10 sec)
• Advantages: Higher energy density, thermal stability, reduced flammability

7. PX4 Autopilot

7.1 Architecture

PX4 follows a microkernel philosophy, using the uORB (Micro Object Request Broker) publish-subscribe bus for inter-module communication.

┌──────────────────────────────────────────────────┐
│                PX4 Flight Stack                   │
│                                                   │
│  Sensor Drivers ──┐                               │
│  EKF2 Estimator ──┼──→ [uORB Bus] ──→ MAVLink    │
│  Control Modules──┤              ──→ Mixer/Actuators│
│  Flight Modes   ──┘              ──→ Logger       │
└──────────────────────────────────────────────────┘

7.2 Flight Modes

Manual Modes:

Assisted Modes:

Autonomous Modes:

7.3 MAVLink Protocol

A lightweight messaging protocol for communication between drones and ground stations or companion computers.

MAVLink v2 Message Structure:

[STX][LEN][INC][CMP][SEQ][SYS_ID][COMP_ID][MSG_ID(3B)][PAYLOAD][CRC][SIGNATURE]

• Heartbeat: Sent at 1Hz, includes vehicle type, flight mode, and system status
• QGroundControl: Open-source ground control station based on MAVLink

8. ArduPilot

8.1 PX4 vs ArduPilot

8.2 Supported Vehicles

9. ROS2 Integration

9.1 Two Integration Approaches

┌──────────────────┐     ┌──────────────────┐
│ Companion Computer│     │ Pixhawk (PX4)    │
│                  │     │                  │
│  [ROS2 Nodes]    │     │  [PX4 Autopilot] │
│       │          │     │       │          │
│  [uXRCE-DDS     │◄───►│  [uXRCE-DDS     │  ← Approach 1 (Recommended)
│   Agent]         │UART │   Client]        │
│       or         │     │                  │
│  [MAVROS]       │◄───►│  [MAVLink]       │  ← Approach 2 (Legacy)
└──────────────────┘     └──────────────────┘

Approach 1: uXRCE-DDS (Recommended)

• PX4's built-in DDS client participates directly in the ROS2 DDS network
• uORB topics map directly to ROS2 topics
• Bypasses MAVLink for lower latency and higher bandwidth

Approach 2: MAVROS (Legacy)

• MAVLink to ROS2 topic/service bridge
• Compatible with ArduPilot

9.2 Offboard Control Example

import rclpy
from rclpy.node import Node
from px4_msgs.msg import (
    OffboardControlMode,
    TrajectorySetpoint,
    VehicleCommand
)

class OffboardControl(Node):
    def __init__(self):
        super().__init__('offboard_control')
        self.offboard_pub = self.create_publisher(
            OffboardControlMode, '/fmu/in/offboard_control_mode', 10)
        self.trajectory_pub = self.create_publisher(
            TrajectorySetpoint, '/fmu/in/trajectory_setpoint', 10)
        self.command_pub = self.create_publisher(
            VehicleCommand, '/fmu/in/vehicle_command', 10)

    def arm(self):
        msg = VehicleCommand()
        msg.command = VehicleCommand.VEHICLE_CMD_COMPONENT_ARM_DISARM
        msg.param1 = 1.0  # Arm
        self.command_pub.publish(msg)

    def set_position(self, x, y, z):
        msg = TrajectorySetpoint()
        msg.position = [x, y, z]  # NED coordinates
        self.trajectory_pub.publish(msg)

10. SITL/HITL Simulation

10.1 SITL (Software-In-The-Loop)

Runs the flight firmware on a development PC without any physical hardware.

# PX4 SITL + Gazebo
cd PX4-Autopilot
make px4_sitl gz_x500

# Micro XRCE-DDS Agent (separate terminal)
MicroXRCEAgent udp4 -p 8888

# ArduPilot SITL + Gazebo
sim_vehicle.py -v ArduCopter -f gazebo-iris --console --map

• Runs the exact same code as real firmware (only simulation drivers differ)
• Risk-free algorithm validation and rapid iterative development

10.2 HITL (Hardware-In-The-Loop)

Runs standard firmware on actual Pixhawk hardware, with the simulator providing sensor data.

• Validates real hardware timing and performance
• Catches hardware compatibility issues early
• Final verification step before real flight

10.3 Gazebo Integration

Gazebo Harmonic (currently recommended):

• Physics engine, sensor simulation, camera/LiDAR rendering
• Communicates with PX4 via MAVLink or direct plugin interface
• Integration with ROS2 Humble

11. Learning Roadmap

11.1 Beginner (1-2 Months)

11.2 Intermediate (2-4 Months)

11.3 Advanced (4-6 Months)

12. References

Official Documentation

1. PX4 User Guide
2. PX4 Dev Guide
3. ArduPilot Copter Docs
4. MAVLink Guide
5. QGroundControl User Guide
6. Pixhawk Open Standards

Hardware

7. Holybro Pixhawk 6X
8. Holybro Pixhawk 6C
9. Dronecode FMUv6 Family

Control and Estimation

10. PX4 Controller Diagrams
11. PX4 EKF2 Tuning Guide
12. ArduPilot EKF Overview
13. Kalman Filter Tutorial

ESC Protocols

14. DShot Protocol - Oscar Liang
15. ESC Firmware and Protocols - Oscar Liang
16. ArduPilot PWM/DShot ESCs

eVTOL

17. eVTOL Guide - Dewesoft
18. eVTOL Market Analysis - MotorWatt
19. FAA eVTOL Certification
20. EHang Solid-State Battery Flight

ROS2 Integration

21. PX4 ROS2 Integration
22. PX4 uXRCE-DDS Bridge
23. MAVROS GitHub

GitHub Repositories

24. PX4/PX4-Autopilot
25. ArduPilot/ardupilot
26. px4-offboard (ROS2 Example)
27. MAVSDK

Quiz