This paper considers the problem of estimating pose from inertial and bearing-only vision measurements. We present a non-linear observer that evolves directly on the special Euclidean group SE(3) from inertial measurements and bearing measurements, such as provided by a visual system tracking known landmarks. Local asymptotic convergence of the observer is proved. The observer is computationally simple and its gains are easy to tune. Simulation results demonstrate robustness to measurement noise and initial conditions.

This paper presents a coupled observer that uses accelerometer, gyrometer and vision sensors to provide estimates of pose and linear velocity for an aerial robotic vehicle. The observer is based on a non-linear complimentary filter framework and incorporates adaptive estimates of measurement bias in gyrometers and accelerometers commonly encountered in low-cost inertial measurement systems. Asymptotic stability of the observer estimates is proved as well as bounded energy of the observer error signals. Experimental data is provided for the proposed filter run on data obtained from an experiment involving a remotely controlled helicopter.