Dead Reckoning for Autonomous Vehicles

Dead Reckoning and Sensor Fusion for Accurate Navigation [Magnetometer Calibration]: • Corrected hard iron distortions by centering the data using ellipse fitting. • Addressed soft iron distortions by scaling the data based on ellipse parameters to normalize axis proportions. • Improved yaw estimation accuracy through these calibrations. [Sensor Fusion for Yaw Estimation]: • Implemented a complementary filter to combine noisy magnetometer data with low-drift gyroscope data. • Applied a low-pass filter to reduce magnetometer noise and a high-pass filter to enhance gyroscope readings. • Achieved reliable yaw estimates by experimenting with weighting parameters in the complementary filter. [Forward Velocity Estimation]: • Corrected accelerometer bias using a convolutional window to adjust for baseline shifts. • Integrated bias-corrected acceleration data to compute forward velocity. • Accounted for stops by detecting and removing drift caused by prolonged idle periods. [Trajectory Analysis]: • Compared GPS-derived trajectories with IMU-based estimates to assess accuracy. • Found good alignment over short intervals but noted drift over longer paths due to accumulated sensor bias. This project effectively combined calibration, filtering, and integration techniques to enhance navigation accuracy and demonstrated the challenges of sensor bias and noise in real-world environments.