Autonomous vehicle localization in a vector field: Underwater vehicle implementation

A global localization method for autonomous underwater vehicles (AUVs) based on background flow velocity map prediction is proposed. Background flow velocity maps with time stamps are assumed to be predicted by a regional ocean model with certain accuracy and these maps are preloaded onto vehicles. Vehicles measure (estimate) general absolute flow velocities at their locations through local sensing. Particle filters are implemented to estimate AUVs' locations by comparing local flow velocity measurements with flow velocity prediction. The proposed algorithm is path independent and is suitable for any non-uniform vector fields. Convergence of localization error is achieved in simulation tests in a time dependent benchmark flow field, namely the double-gyre flow field. A two-dimensional vector field interpolation scheme based on irregularly-spaced data points is implemented to estimate theoretical flow velocities at vehicles' locations. It is shown that interpolation error can be treated as low-level map noise, which has limited impact on convergence. Localization error converges in simulation tests with different levels of map noise, which we believe to be dominant in practice comparing to measurement noise and vehicle control noise.