State Estimation for Autonomous Vehicles

As the Lead of the State Estimation Team at aUToronto, I led the development of multi-sensor fusion algorithms using an Extended Kalman Filter (EKF). This system integrated data from GPS, IMUs, wheel encoders, LiDARs and cameras to ensure accurate localization, even under sensor failure conditions.

To test and enhance the system, I implemented a variable L-Band attenuator to simulate GPS signal degradation during vehicle testing. This allowed us to evaluate the EKF's robustness and dynamically pivot between sensors using integrity monitoring. The system leveraged a chi-squared test to analyze residuals and validate sensor reliability.

I also designed a custom bias-tee PCB to solve power supply issues in the GPS system. This solution enabled simultaneous RF signal and DC current transmission, ensuring reliable GPS operation during testing. Additionally, I developed a Map Offset GUI using PyQt to correct positional and heading errors, reducing localization inaccuracies by 87%.

Through rigorous testing and iteration with Google Test and GMock, I helped lead to our team to secure 1st place in all events at the SAE AutoDrive Challenge.