Best IMU Drift Correction Planning & Solution

Pitch and roll (attitude) vectors have an absolute global physical reference: Earth’s gravitational acceleration vector (1G pointing straight down). Accelerometers measure this vector continuously during stationary or quasi-static states to instantly bound pitch and roll tilt errors. However, yaw (heading) rotates perpendicular to gravity, meaning accelerometers provide zero rotational constraint around the Z-axis. Without magnetometer data or external optical references, yaw relies exclusively on the numerical integration of the gyroscope rate vector, which drifts steadily over time due to sensor bias.

In double-integration position tracking (p = ∫∫ a dt²), any tiny acceleration offset integrates into linear velocity error (v = a·t), and subsequently into quadratic position error (p = 0.5·a·t²). During human or legged locomotion, each foot remains completely motionless on the ground for approximately 20% to 40% of the gait cycle (the stance phase). ZUPT detects this static phase using threshold detectors and artificially feeds a velocity measurement of exactly 0.0 m/s into the Kalman filter. This forces the velocity error state to zero, stopping error accumulation in its tracks.

Yes. In outdoor environments free from reinforced steel structures, magnetometers operate under pristine geomagnetic conditions, allowing the 9-axis fusion algorithm to lock yaw indefinitely with zero static drift. By combining pure geomagnetic heading with biomechanical kinematic link solvers and ground contact ZUPT, full-body performers or humanoid robots can navigate large open terrains while keeping body drift tightly bounded.

Allan Variance is an offline stochastic identification technique used during R&D and factory calibration to quantify the exact noise profile of the sensor (including Angle Random Walk, Bias Instability, and Rate Random Walk). The parameters derived from Allan Variance curves are then programmed into the Extended Kalman Filter’s process noise covariance matrix Q. Runtime Kalman filtering is the online real-time state observer that uses those offline parameters to optimally balance sensor confidence frame by frame.

Virdyn employs adaptive magnetic disturbance rejection (AMDR). The software continuously calculates the magnitude and dip angle of the received magnetic field and compares them against the known local Earth magnetic model. If an anomaly is identified (e.g., stepping near a steel truss), the Kalman filter automatically reduces the measurement confidence of the magnetometer to zero, preserving orientation purely via high-speed, temperature-compensated gyroscope integration until the actor exits the magnetic disturbance zone.