Inertial Measurement Unit (IMU)

An accelerometer measures proper acceleration—the rate of change of velocity relative to a free-fall (inertial) reference frame—along one or more axes. In an IMU, a triaxial (3-axis) accelerometer measures specific force, which is the sum of kinematic acceleration and the reaction force due to gravity.

• Mechanism: Most MEMS (Micro-Electro-Mechanical Systems) accelerometers use a tiny proof mass suspended by springs; acceleration causes displacement, which is measured capacitively.
• Primary Output: Linear acceleration in m/s² along the X, Y, and Z axes.
• Critical Limitation: It cannot distinguish between kinematic motion and the constant pull of gravity. A stationary IMU on a flat surface will report a +9.81 m/s² acceleration along the axis pointing upward, opposite gravity.

A gyroscope (or gyro) measures angular velocity—the rate of rotation around an axis—in degrees per second (°/s) or radians per second (rad/s).

• Mechanism: Modern MEMS gyroscopes typically use the Coriolis effect: a vibrating proof mass experiences a force proportional to the rotation rate, which is measured as a capacitance change.
• Primary Output: Angular rate around the X (roll), Y (pitch), and Z (yaw) axes.
• Critical Limitation: Gyroscopes suffer from drift (bias instability) and scale factor errors. Even tiny, uncorrected biases integrate over time into massive errors in estimated orientation, a fundamental challenge in inertial navigation.

A magnetometer measures the strength and direction of the ambient magnetic field, typically the Earth's magnetic field, to provide an absolute heading reference. While not strictly required for an IMU's inertial function, it is a common addition in Attitude and Heading Reference Systems (AHRS).

• Mechanism: Often a Hall-effect or magnetoresistive sensor that detects changes in magnetic flux density.
• Primary Output: Magnetic field vector in microteslas (µT) along three axes.
• Critical Limitation: Highly susceptible to hard iron (permanent magnets) and soft iron (ferromagnetic materials causing field distortion) interference from the local environment, such as motors, steel structures, or electronic devices.

The raw, noisy, and drifting outputs from the individual sensors are useless for navigation without a sensor fusion algorithm. This is the computational core that combines the data streams into a stable, accurate estimate of orientation, and sometimes position and velocity.

• Primary Function: Fuses high-frequency, short-term accuracy of gyroscopes with the long-term, drift-free reference of accelerometers (for gravity vector) and magnetometers (for North).
• Common Algorithms:
  • Complementary Filter: A simple, computationally efficient high-pass/low-pass filter blend of gyro and accelerometer data.
  • Kalman Filter (or Extended Kalman Filter): The industry-standard optimal recursive estimator that models sensor noise and system dynamics to produce a statistically best guess of the true state.
• Output: A fused 6-DoF or 9-DoF orientation quaternion or Euler angles (roll, pitch, yaw).

When an IMU is paired with an initial position and sophisticated dead reckoning algorithms, it becomes the heart of an Inertial Navigation System. An INS integrates acceleration twice to estimate velocity and position, and integrates angular rate to estimate orientation.

• Process:
  1. Attitude Update: Gyro data updates the orientation quaternion.
  2. Coordinate Transformation: Measured specific force (from accelerometer) is rotated from the body frame to the global navigation frame using the current orientation.
  3. Gravity Subtraction: The known gravity vector is subtracted to isolate kinematic acceleration.
  4. Integration: Acceleration is integrated once for velocity, and twice for position.
• Critical Challenge: Cubic error growth. Any uncorrected bias in the accelerometer leads to an error in position that grows with the cube of time (∝ t³). This is why INS is almost always fused with GPS or other external aids in a loosely or tightly coupled architecture.

IMU performance is quantified by specific, measurable parameters that directly impact system accuracy and cost.

• Bias Instability: The limit of the gyroscope's or accelerometer's bias drift over time, measured in °/hr or µg. Lower is better. Defines the minimum orientation drift.
• Angle Random Walk (ARW): The noise floor of a gyroscope, causing angle error that grows with the square root of time. Measured in °/√hr.
• Velocity Random Walk (VRW): The noise floor of an accelerometer, causing velocity error that grows with the square root of time. Measured in m/s/√hr.
• Scale Factor Error: The deviation of the sensor's output from its ideal, linear response to input. Expressed as a percentage (%).
• Non-Orthogonality: The error in alignment between the sensor's axes, measured in degrees or milliradians.
• Bandwidth: The frequency range over which the sensor provides an accurate measurement, critical for capturing dynamic motion.

The algorithmic process of combining data from multiple disparate sensors (like an IMU, camera, and LiDAR) to produce a more accurate, complete, and reliable state estimate than is possible with any single sensor alone.

• Key Techniques: Include filtering (Kalman Filter), smoothing, and deep learning-based fusion.
• Purpose: Mitigates the weaknesses of individual sensors. For example, an IMU's high-frequency data can correct for motion blur in a camera, while the camera's absolute positioning can correct for the IMU's drift.

A specific and critical sensor fusion technique that tightly couples visual data from a camera with inertial data from an IMU to estimate the 6-degree-of-freedom (6DOF) pose and velocity of a robot or vehicle.

• Core Advantage: Provides robust, metric-scale pose estimation where pure visual odometry (VO) fails (e.g., during rapid motion, textureless environments, or temporary occlusion).
• Output: A continuous stream of position, orientation, and velocity, which is foundational for navigation and mapping (SLAM).

An optimal recursive algorithm for estimating the state of a linear dynamic system from a series of noisy measurements. It operates in a two-step cycle: prediction (using a motion model) and update (correcting with a new sensor measurement).

• Application to IMUs: The Kalman filter's prediction step is often driven by the IMU's kinematic model. The update step can incorporate corrections from GPS, wheel encoders, or visual features.
• Foundation: It is the mathematical bedrock for most real-time sensor fusion systems involving IMUs.

The nonlinear version of the Kalman Filter. It linearizes the system's nonlinear dynamics and measurement models around the current state estimate, making it the workhorse algorithm for fusing IMU data in real-world robotics.

• Why it's necessary: IMU and camera models are inherently nonlinear. The EKF handles these nonlinearities to perform state estimation for problems like Visual Inertial Odometry (VIO) and SLAM.
• Process: At each step, it computes a local linear approximation (Jacobian) of the system, then applies the standard Kalman filter equations.

The computational problem of constructing or updating a map of an unknown environment while simultaneously tracking the agent's location within it. IMUs are a standard component in modern SLAM pipelines.

• Role of the IMU: Provides high-frequency ego-motion estimates between camera or LiDAR frames, making the data association and optimization problems more robust and efficient.
• Types: Visual-Inertial SLAM (VI-SLAM) and LiDAR-Inertial SLAM are common architectures that leverage IMUs to improve accuracy and robustness.

An operating system designed for deterministic, time-critical applications where sensor processing and control must occur within strict, predictable time constraints—essential for reliable IMU data handling.

• Importance for IMUs: Guarantees that the high-frequency interrupt from an IMU is serviced with minimal and bounded latency (jitter), ensuring sensor data is processed and fused in a timely manner for stable control loops.
• Examples: FreeRTOS, Zephyr, and VxWorks are commonly used in embedded robotic and automotive systems.