Distributed simulations have become valuable tools for individual and group training. A combination of live, virtual and constructive distributed simulations that is highly promising for greater realism in training at reduced costs, called embedded simulation, is being explored by the U. S. Army's Simulation, Training and Instrumentation Command (STRICOM) Inter-Vehicle Embedded Simulation Technology (INVEST) Science and Technology Objective (STO) program for use in combat vehicles. Among the many technical challenges to be overcome is that of providing a simulation environment in which live vehicles, manned vehicle simulators, and computer generated forces can interact with each other as well as with the battlefield environment in real-time over a geographically diverse, distributed network. The main problem is the high communications requirements imposed by the need to convey large amounts of data among the various players. The Vehicle Model Generation and Optimization for Embedded Simulation (VMGOES) project at the University of Central Florida is focusing on this aspect of the INVEST STO program. The approach is to use a behavioral vehicle model that is context-based to match the actions of the human-controlled entity on the battlefield. By observing the surrounding environment of the vehicle model's location in the simulation at each update time step, the model will determine what context it should be in and perform the actions that are appropriate for that context. This will allow the vehicle model to match the human-controlled entity's behavior for a longer period of time than is possible with only dead-reckoning updates, thus reducing the communications bandwidth required. However, discrepancies between the vehicle model and the human controlled entity will inevitably occur and these must be detected and resolved to allow the vehicle model to function efficiently. The portion of our model that addresses this need, the Difference Analysis Engine (DAE), will be resident on the human-controlled entity. It will be able to observe the actual vehicle's actions as well as the simulation environment and the vehicle model itself. It then must evaluate whether significant discrepancies exist. If they do, it will immediately take the action needed to synchronize the vehicle model with the actual entity. These corrections can involve a simple State Realignment to update the vehicle model's location, direction and speed; a forced vehicle model Context Shift to match the context of the human-controlled entity; a Model Correction to change the way the model itself responds; and, as a last resort, a Model Suspension to revert to standard dead-reckoning until the DAE can recognize what context the human-controlled vehicle actually is in. This paper will focus on those DAE functions and on how techniques, such as temporal template based reasoning, neural networks and genetic algorithms, are being used to accomplish those DAE functions.