Project
Autonomous Vehicle Navigation
Michigan Autonomous Robotic Vehicle Team · Sep 2026 – Present
Overview
I work on the navigation subteam of the Michigan Autonomous Robotic Vehicle Team.
MPPI Path Tracking Controller
I develop a Model Predictive Path Integral (MPPI) controller and its dynamics and cost functions. I use vectorized PyTorch calculations to project sampled robot positions onto path segments to calculate the distance from the path to the robot and find the closest point.
Chord-Length Odometry
Replacing a midpoint approximation with chord-length-based odometry, reducing the root-mean-square error by about 80% across simulation runs. Essentially, the previous method stretches out the arc distance and points that entire length as a straight line towards the mid heading, which overshoots if the robot is turning. The chord-length method instead uses the length of the chord between the two points. See the PR for a complete description of the math/geometry, implementation, testing, and results