Localization

My work has included ROS2/C++ EKF localization using odometry and cone detections. This approach combines sensor information to estimate a consistent vehicle state while handling noisy measurements and imperfect observations.

Robustness is important in this setting because cone detections can be unreliable and noisy. The system needs to remain stable under imperfect inputs and maintain dependable state estimation during operation.

SLAM

I have also worked with GTSAM and iSAM2 for graph-based SLAM, including data association and loop closure. This was used to build a more globally consistent map and improve localization quality over longer runs.

The work has included offline evaluation of SLAM performance, with ongoing effort toward reliable online execution in a real racing stack.

Trajectory optimization

My planning work includes B-spline trajectory representation and minimum-curvature optimization. This formulation is useful for generating smooth, drivable paths that respect the vehicle dynamics and the constraints of the environment.

This optimization is integrated with CarMaker in Python, providing an environment for evaluating the planned path in a realistic simulation workflow.

Current engineering focus

The immediate engineering focus is on robustness to noisy and outlier cone detections, thorough testing, reliable execution, and continued work on online localization and SLAM. There is also active attention to midline generation and producing dependable trajectory behavior in dynamic, uncertain conditions.