A pattern-based modeling framework for simulating human-like pedestrian steering behaviors

n this paper, we propose a new approach to modeling natural steering behaviors of virtual humans. We suspect that a small number of steering strategies are sufficient for generating typical pedestrian behaviors observed in daily-life situations. Through these limited strategies we show that complex steering behaviors are generated by executing appropriate steering strategies at the appropriate time. In our model, decisions on the selection, scheduling and execution of steering strategies in a given situation are based on the matching results between the currently perceived spatial-temporal patterns and the prototypical cases in an agent's experience base. From a modeler's point of view, our approach is intuitive to use. Our model is carefully evaluated through a three-stage validation process, using experimental studies on basic test scenarios, model comparisons under standard but more complex test scenarios, and sensitivity analysis on key model parameters. Experimental results show that our model is able to generate results that reflect the collective efficiency of crowd dynamics and is in agreement with existing literature on pedestrian studies.