Active Model Discrimination Using Partition-Based Output Feedback Input Design

In this paper, we propose a partition-based output feedback active model discrimination approach that generates optimal output feedback inputs in a fixed time horizon for separating a set of discrete-time affine models subject to uncontrolled inputs, noises and uncertain initial conditions. Instead of computing the optimal input by solving a parametric mixed-integer linear program (MILP) at run time, we move this computationally demanding optimization task offline by partitioning the measurement domain and building a partition tree over the fixed time horizon. Since output measurements are available at each time instant during run time, we can update the separating input correspondingly and improve the model discrimination performance by reducing the input cost. The effectiveness of the proposed approach is demonstrated through simulations for identifying intention models of human-driven vehicles in a lane changing scenario.