We describe a real-time, low-cost system to localize a mobile robot in outdoor environments. Our system relies on stereo vision to robustly estimate frame-to-frame motion in real time (also known as visual odometry). The motion estimation problem is formulated efficiently in the disparity space and results in accurate and robust estimates of the motion even for a small-baseline configuration. Our system uses inertial measurements to fill in motion estimates when visual odometry fails. This incremental motion is then fused with a low-cost GPS sensor using a Kalman Filter to prevent long-term drifts. Experimental results are presented for outdoor localization in moderately sized environments (\geqslant 100 meters)