This poster describes ongoing work to use stereo vision for robotic boat navigation. Stereo is low-power which makes it attractive for embedded applications. Stereo is also a challenge due to the complexity of the application domain, the myriad effects of varying illumination, and inherent bias and calibration errors. We describe a preliminary design for our system that will enable an autonomous surface vehicle to navigate using a waypoint following mode through the fusion of measurements from stereo vision, inertial sensors and GPS. We describe initial results obtained from lab tests and field experiments at Lake Fulmor.