| Embedded simulations hold great promise for highly realistic training but they have technical challenges to overcome. One of those, the limitation on available bandwidth to meet the huge communication demands of embedded simulation, is being addressed through research on the use of behavioral models. These models are intended to improve on current dead reckoning modeling techniques to reduce the communications needed. This is accomplished by synchronizing the behavioral models of human-controlled vehicles with the actual vehicle such that the models can be accurate for longer periods of time. The synchronization is done through the use of a hierarchical, context-based representation, whereby the behavioral model, located on the other vehicles in an embedded simulation, performs the actions that are appropriate for the behavioral context and sub-contexts of the actual vehicle it represents. However, the model has to know what the current behavioral context of the human-controlled vehicle is in order to respond with the correct actions. This dissertation focuses on the difficult problem of recognizing the behavioral context in real-time and then synchronizing the distributed behavioral models with the actions of the human-controlled vehicle. Hierarchical, template-based reasoning is used as the basis for the behavior recognition, where templates represent each behavioral context or sub-context. The weight given each template is critical to the correct selection of the template that identifies the current behavior and is based on weighted attributes of the vehicle's state and its surrounding environment. Only manual setting of those attribute weights have been done before and would be prohibitively time-consuming, if at all possible, for this domain. Therefore, this research develops and implements a novel learning by observation methodology using fuzzy membership sets and neural networks to automate the setting of template attribute weights to allow significant discrimination between different categorized behavioral contexts/sub-contexts on the human-controlled vehicle. |
