GPS & IMU Sensor Fusion for Automotive Dead Reckoning

Navigating a vehicle through dense urban environments—where GPS signals flicker between skyscrapers and IMU sensors drift like a compass in a storm—is a formidable challenge. This project engineered a sensor fusion pipeline to merge the IMU’s rapid motion tracking with GPS’s steady geospatial anchors, delivering a robust solution for autonomous navigation in signal-degraded areas.

Technical Approach

1. Sensor Setup & Calibration:

   • VN-100 IMU: Captured 100 Hz accelerometer, gyroscope, and magnetometer data, mounted on the vehicle dashboard.

   • BU-353S4 GPS: Provided 1 Hz latitude/longitude fixes via NMEA GPGGA strings, roof-mounted for optimal sky visibility.

   • Magnetometer Calibration:

     • Hard Iron Correction: Removed static magnetic offsets caused by the vehicle’s metal frame, recentering raw data to the origin.

     • Soft Iron Correction: Normalized elliptical distortions in the magnetic field using axis scaling, achieving a near-perfect unit-circle distribution.

2. Yaw Estimation via Complementary Filtering:

   • Magnetometer Yaw: Derived from tilt-compensated magnetic field measurements, stable but prone to urban electromagnetic interference.

   • Gyroscope Yaw: Integrated angular velocity for smooth short-term tracking but susceptible to drift.

   • Fusion Strategy: A complementary filter blended the two:

     • Low-Pass Filter (0.15 Hz): Smoothed magnetometer noise (e.g., power lines, steel structures).

     • High-Pass Filter (2 Hz): Suppressed gyroscope drift, preserving agility during sharp turns.

   • Outcome: 42% reduction in yaw error compared to raw gyroscope integration.

3. Velocity Alignment & Trajectory Reconstruction:

   • IMU Velocity: Bias-corrected accelerometer data was integrated to estimate speed, with high-pass filtering (0.1 Hz) to suppress drift.

   • GPS Validation: Velocity derived from Haversine-calculated distances between consecutive GPS fixes, acting as ground truth during stops.

   • Trajectory Estimation:

     • Decomposed IMU velocity into Easting/Northing components using fused yaw.

     • Integrated velocities over time to reconstruct the vehicle’s path, aligning it with GPS coordinates via initial heading rotation.

Key Results

• Heading Stability: Complementary filter reduced yaw drift by 42%, critical for lane-keeping in urban turns.

• Velocity Consistency: IMU-GPS speed estimates aligned within 0.48 m/s RMSE post-correction.

• Trajectory Precision: IMU dead reckoning maintained <2 m deviation from GPS for 90 seconds—long enough to navigate short urban blocks or tunnels.

Why It Matters

This project bridges the gap between high-frequency sensor agility and low-frequency geospatial reliability, proving that autonomous systems can thrive even when GPS falters. By mastering sensor fusion, calibration, and real-time filtering, the system:

• Mitigated urban magnetic distortions.

• Delivered sub-meter velocity accuracy.

• Extended reliable navigation during GPS outages.

Skills Demonstrated

• Sensor Fusion: Harmonized IMU and GPS data streams using complementary filtering.

• Algorithm Design: Implemented calibration, filtering, and integration workflows in MATLAB.

• ROS Integration: Synchronized multi-sensor data with ROSbags for cohesive analysis.

• Geospatial Analytics: Transformed raw GPS coordinates into actionable trajectories.

Impact & Future Vision

This work is a blueprint for urban-ready autonomy, addressing real-world challenges like signal blockages and sensor drift. Future iterations could integrate Kalman filters for adaptive noise handling or fuse LiDAR data for SLAM-based navigation.