SKASai Kiran Athaluri

← All projects

RKCRobotics, Kinematics & ControlNov – Dec 2024

GPS & IMU Sensor Fusion for Automotive Dead Reckoning

Custom Python ROS drivers characterize sensor noise via Allan Variance and fuse magnetometer + gyro yaw with a complementary filter; a causal “moving-frame” zero-velocity update compensates IMU drift, and kinematics (accounting for sensor offset from the center of mass) reconstruct the trajectory against GPS ground truth.

  • ROS
  • Python
  • Allan Variance
  • Complementary Filter
  • Sensor Fusion
  • MATLAB

↗ gitlab.com/athaluri.s

Hardware & Architecture

  • Vehicle-mounted GPS + IMU modules acquired over a ROS framework with self-written Python device drivers.
  • Three datasets: circular roundabout driving (magnetometer calibration), a 5-minute stationary hold (bias), and a looped main-road drive (validation).

Key Highlights

  • Hard-iron correction by mean subtraction produced a near-spherical magnetic flux plot, confirming negligible soft-iron distortion.
  • Complementary filter fuses high-pass gyro yaw with low-pass magnetometer yaw to eliminate low-frequency drift.
  • Causal moving-frame windowing detects rest via low std-dev and zeroes velocity, offsetting integration bias in real time.
  • Kinematic model accounts for the sensor's ~4 cm displacement from the center of mass; dead-reckoned path tracks the GPS ground truth.

Images

Corrected Magnetic Flux (X vs Y). The spherical nature of the post-calibration plot verifies the successful removal of hard-iron offsets and the absence of soft-iron distortion.
Comparison of raw Gyroscope Yaw, raw Magnetometer Yaw, and the fused Complementary Filter output, demonstrating the elimination of low-frequency drift.
Velocity vs Time graph illustrating the raw IMU integration drift compared against the causal 'Moving Frame' zero-velocity update algorithm and ground-truth GPS.
Calculated East Heading vs North Heading (Dead Reckoning) plotted against the GPS reference path, showcasing the kinematic tracking capabilities and scale adjustments.

Code

The full ROS device drivers plus the magnetometer-calibration, complementary-filter, and dead-reckoning analysis scripts live on the linked GitLab profile.