This paper presents and evaluates a relative localization scheme for a heterogeneous team of low-cost mobile robots. An error-state, complementary Kalman Filter was developed to fuse analytically-derived uncertainty of stereoscopic pose measurements of an aerial robot, made by a ground robot, with the inertial/visual proprioceptive measurements of both robots. Results show that the sources of error, image quantization, asynchronous sensors, and a non-stationary bias, were sufficiently modeled to estimate the pose of the aerial robot. In both simulation and experiments, we demonstrate the proposed methodology with a heterogeneous robot team, consisting of a UAV and a UGV tasked with collaboratively localizing themselves while avoiding obstacles in an unknown environment. The team is able to identify a goal location and obstacles in the environment and plan a path for the UGV to the goal location. The results demonstrate localization accuracies of 2cm to 4cm, on average, while the robots operate at a distance from each-other between 1m and 4m.