STOCHASTIC SIMILARITY FOR VALIDATING HUMAN CONTROL STRATEGY MODELS

Modeling dynamic human control strategy (HCS), or human skill in respo nse to real-time sensing is becoming an increasingly popular paradigm in many different research areas, such as intelligent vehicle systems, virtual reality, and space robotics. Such models are often learned fr om experimental data, and as such can be characterized despite the lac k of a good physical model. Unfortunately, learned models presently of fer few, if any, guarantees in terms of model fidelity to the training data, This is especially true for dynamic reaction skills, where erro rs can feed back on themselves to generate state and command trajector ies uncharacteristic of the source process, Thus, we propose a stochas tic similarity measure-based on hidden Markov model (HMM) analysis-cap able of comparing and contrasting stochastic, dynamic, multidimensiona l trajectories. This similarity measure is the first step in validatin g a learned model's fidelity to its training data by comparing the mod el's dynamic trajectories in the feedback loop to the human's dynamic trajectories. In this paper, we first derive and demonstrate propertie s of the similarity measure for stochastic systems, We then apply the similarity measure to real-time human driving data by comparing differ ent control strategies among different individuals. We show that the p roposed similarity measure out performs the more traditional Bayes cla ssifier in correctly grouping driving data from the same individual, F inally, we illustrate how the similarity measure can be used in the va lidation of models which are learned from experimental data, and how w e can connect model validation and model learning to iteratively impro ve our models of human control strategy.