Partition-Based Parametric Active Model Discrimination with Applications to Driver Intention Estimation

In this paper, we propose a partition-based parametric active model discrimination approach that distinguishes among a set of discrete-time affine time-invariant models with uncontrolled inputs, model-independent parameters that are revealed in real-time and noise. By partitioning the operating region of the parameters, the problem turns into a sequence of offline optimization problems. Thus, at each time instant, we only need to determine which subregion in the resulting partition tree the revealed parameters lie in and select the corresponding pre-computed inputs. The offline optimal input design problem is formulated as a bilevel problem and further cast as a mixed-integer linear program (MILP). Finally, we demonstrate the effectiveness of the proposed approach for estimating driver intention in a lane-changing scenario.