Sensor Calibration

Intrinsic calibration determines the internal parameters of a sensor that define its geometric and optical properties. For cameras, this involves finding the focal length, principal point, and lens distortion coefficients. For an Inertial Measurement Unit (IMU), it involves characterizing scale factors, biases, and non-orthogonalities of its accelerometers and gyroscopes. This process corrects for imperfections inherent to the sensor's manufacturing.

• Example: Using a checkerboard pattern to calibrate a camera's pinhole model and radial distortion.
• Goal: To enable accurate metric measurements from the sensor's raw output.

Extrinsic calibration determines the spatial relationship—translation and rotation—between different sensors or between a sensor and a robot's base frame. This establishes the coordinate transform needed to fuse data from multiple sources into a common reference frame.

• Example: Hand-eye calibration finds the transform between a robot's end-effector and a wrist-mounted camera.
• Example: Calibrating the rigid transform between a LiDAR and an IMU for a self-driving car's perception stack.
• Result: A transformation matrix (e.g., 4x4 homogeneous matrix) that maps points from one coordinate system to another.

Temporal calibration synchronizes the timestamps of data streams from different sensors. It determines the fixed time offset and, if necessary, the clock drift between devices. This is critical for sensor fusion algorithms (e.g., Kalman filters) that assume measurements are aligned in time.

• Cause: Independent internal clocks, varying sensor processing delays, and network latency.
• Method: Often performed by detecting a synchronous event (e.g., a flashing LED captured by a camera and photodiode) or using hardware triggers and protocols like Precision Time Protocol (PTP).
• Impact: Mis-synchronization of just milliseconds can cause significant state estimation errors in fast-moving systems.

Inter-sensor calibration is a holistic process that jointly optimizes the parameters of multiple, different sensors. It addresses the coupling between intrinsic, extrinsic, and temporal parameters that can occur when calibrating sensors separately introduces correlated errors.

• Use Case: Calibrating a camera-LiDAR-IMU suite simultaneously, where the camera's focal length estimate can affect the estimated translation to the LiDAR.
• Advantage: Produces a globally consistent set of parameters, often yielding higher overall system accuracy than sequential calibration.
• Complexity: Requires sophisticated optimization over a high-dimensional parameter space and rich, multi-modal calibration data.

Online calibration algorithms run during normal system operation to adapt to changing conditions. They correct for parameters that drift over time due to temperature changes, mechanical stress, or aging components.

• Methods: Often uses state estimation techniques (e.g., extending a SLAM filter's state vector to include calibration parameters) or observes motion constraints (e.g., zero-velocity updates for IMU bias estimation).
• Example: An autonomous vehicle continuously estimating its camera's focal length and IMU biases during navigation.
• Challenge: Requires careful observability analysis to ensure parameters can be identified without dedicated calibration routines.

This distinction defines the reference standard used for calibration.

• Absolute Calibration: References a known, traceable physical standard. The goal is to produce measurements in absolute, real-world units (e.g., meters, degrees Celsius).
  • Example: Using a NIST-traceable weight to calibrate a force sensor's output in Newtons.
• Relative Calibration: References another sensor or an internal consistency constraint. The goal is to ensure consistency between sensors or measurements, not necessarily absolute accuracy.
  • Example: Calibrating two cameras relative to each other for stereo vision without knowing the exact size of the calibration target.
  • Example: Performing a gyro-compass alignment using the Earth's gravity and rotation vectors to find an IMU's orientation relative to the world frame.

A specific calibration process that determines the precise spatial transformation (a 6-DOF pose) between a robot's end-effector (the 'hand') and a camera mounted on it (the 'eye'). This transformation is critical for tasks where the robot must use visual feedback to guide its manipulation, such as visual servoing or bin picking. The process typically involves moving the robot to multiple poses while observing a calibration pattern (like a checkerboard) with the camera to solve for the unknown transform.

A validation technique where a physical hardware component (e.g., a robot's motor controller or embedded computer) is connected to a real-time simulation of its operational environment. The simulator provides realistic sensor inputs (like synthetic LiDAR point clouds) and receives the hardware's actuator commands. This allows for rigorous, safe, and repeatable testing of hardware logic and firmware before full system integration. It is a cornerstone of functional safety validation.

The algorithmic process of combining data from multiple, often heterogeneous, sensors (e.g., IMU, camera, LiDAR, wheel encoders) to produce a more accurate, complete, and reliable estimate of the system's state than any single sensor could provide. Calibration is a prerequisite for effective fusion, as it ensures each sensor's data is in a common, accurate coordinate frame. Common fusion techniques include the Kalman filter and its non-linear variants (Extended Kalman Filter, Unscented Kalman Filter).

An XML-based file format used extensively in the Robot Operating System (ROS) ecosystem to define a robot's physical properties. A URDF file describes:

• Kinematic tree (links and joints)
• Visual and collision geometry (meshes, simple shapes)
• Inertial properties (mass, center of mass, inertia tensor)
• Sensor and actuator mounting frames Accurate URDF models, which rely on calibrated physical measurements, are essential for simulation (physics-based robotic simulation), motion planning, and control.

A system property where a given set of inputs and initial conditions will always produce the same output within a bounded and predictable time frame. This is non-negotiable for real-time robotic control loops that depend on sensor data. Calibration routines themselves must often be deterministic, and the resulting calibrated sensors feed into control systems (often running on a Real-Time Operating System) that require deterministic data flow to guarantee stability and safety.

A systems engineering discipline focused on identifying when a component, like a sensor, is operating outside its normal parameters (fault detection) and determining the root cause (diagnostics). Calibration establishes the sensor's baseline performance. FDI systems monitor for deviations from this baseline, such as:

• Bias drift (a calibration parameter changing over time)
• Increased noise
• Complete sensor failure Effective FDI enables graceful degradation or safe shutdown of the robotic system.