This document describes using a modified Kalman filter to estimate vehicle velocity from noisy acceleration and velocity sensor measurements. A standard Kalman filter can cause drift over time. The modified filter automatically converges to eliminate drift without requiring characterization of sensor offsets. It uses a single state variable rather than separate variables for position, velocity, etc. Simulation results show the modified filter estimates velocity more accurately with no drift compared to a standard Kalman filter. The modified filter also has advantages in easier computation, implementation on lower-power processors, and systematic design of gains.