BBN Technologies

The quality of training with simulations can depend on the quality of the synthetic humans in the simulation. In its pioneering SIMNET training environment, BBN first used human-performance models in a simulation as synthetic opposition forces.*

More recently, BBN has adapted its human-performance-modeling technology to research, some of which we present below. We stand ready to apply this additional expertise to immersive-learning environments that require high-fidelity modeling of human behavior.

This BBN research has been in the forefront of both the design of cognitive architectures for human performance models and in the development of simulation frameworks for the construction of human performance models and for their use in a broad range of applications. In conjunction with the modeling of human performance, BBN has also addressed the difficult problems of the evaluation and validation of human performance models.

Technology and Applications for Human Performance Models

BBN's Distributed Operator Model Architecture (D-OMAR) provides a suite of OpenSource software tools that enable the instantiation of a cognitive architecture and the development of human performance models. As a general purpose simulation environment, D-OMAR facilitates the development of scenarios that include models for all the entities with which the human performance models interact. When running in real time, the D-OMAR human performance models can interact with human players. When human players are not involved, fast-time operation can be used to shorten run times for scenario trials and speed the development of the models.

The D-OMAR modeling environment has been used in a series of research projects for the NASA Ames and Langley Research Centers and the Air Force Research Laboratory. The studies have examined the design of new equipment and procedures for the flight decks of commercial aircraft, air traffic control centers, and workstations for unmanned aerial vehicle (UAV) operations. The D-OMAR modeling framework has also been used to support research in understanding the sources of human error in these complex work environments.

Simulated conversations are part of this Air-traffic-control research (click here for larger view).

Evaluating the Veracity of Human Performance Models: AMBR

Agent-based Modeling and Behavior Representation (AMBR) — In the AMBR study, BBN established a principled approach to the comparison, evaluation, and validation of models from four state-of-the-art human performance modeling systems. Employing the D-OMAR simulation framework, the four human performance models and human subjects were presented with a series of challenge problems. The analysis methodology with established metrics governed the comparison of the behaviors of the computer models to each other and to the behaviors of human subjects performing the identical tasks. A complete report on the AMBR study has been assembled in book form in Modeling Human Behavior with Integrated Architectures, edited by Kevin A. Gluck and Richard W. Pew.

Optimizing for Realistic Characters

Humans intuitively optimize when making judgements about what course of action to take in a given situation. The better a synthetic character can optimize when making judgements, the more lifelike and accurate the simulation, and the better the training. This is especially important when the synthetic character is an opponent. BBN has a great deal of experience with optimization in behavor simulation, including:

• Advocates and Critics for Tactical Behaviors (ACTB) for Tactical Navigation Planning - A real-time controller based upon a behavior-based genetic algorithm that analyzes multiple courses of action and adapts both high-level strategy and immediate tactical actions to the current situation. We have applied ACTB to the problem of controlling one or more simulated unmanned ground vehicles (UGVs) against multiple hostile UGVs for achievement of higher-level mission goals (e.g., reconnaissance, surveillance, and target acquisition) despite changing environmental conditions, evolving mission requirements, and the need to coordinate multiple entities. Also, we have applied ACTB to several simulation environments, including an Army Research Lab (ARL) simulator that uses the same control language as ARL's real-life UGVs, and a BBN-developed simulator based upon BBN's OpenMap technology, which enables the importation of real world terrain data.
• Adaptive control of synthethic characters in computer games - BBN has integrated and enhanced our ACTB technology for controlling multiple characters within a computer game. The characters exhibit distinct behavior modes depending upon the situation, and adapt their actions based on their current situation and their behavior priorities.
• Automated learning of behaviors - Applying automated mechanisms to capture the behaviors exhibited by humans within a simulated environment, and then enhancing the current behaviors of a simulated entity to demonstrate improved capabilities.

*Synthetic opposition forces also have been called Computer Generated Forces (CGF) and Semi-Automated Forces (SAF), and have become more realistic over the years.