SBIR-STTR Award

State-of-the-art Semi-Automated Forces (SAFs) are predictable,non-adaptive, behaviorally distinguishable from manned simulators, and once"figured out", the only remedy for these performance shortcomings is acostly rewrite of the behavioral model software. We propose a hybrid SAFcontrol architecture that integrates neural networks with rule-basedsystems to enable human like learning, allowing a SAF system tocontinuously learn from manned simulators, human instructors, and its ownmistakes. This technology will make SAF behaviors more realistic, and willsave time and money by permitting Battle Trainers to modify SAF behaviorwithout the intervention of programmers. The development team includesexperts in robotic skill acquisition and neural network-based control, aswell as some of the original developers of SIMNET SAF. Phase I willdetermine how to integrate concepts of hybrid cooperative and reinforcementlearning into an existing SAF system, demonstrate behavior augmentation andperformance gains attributable to SAF learning for two battle scenarios,and propose a method for integrating such learning into state-of-the-artModSAF and Distributed Interactive Simulation.

In order to keep pace with events in a changing world and properly prepare dismounted infantry for operations in urban environments, there exists a need for an easy-to-use CGF development system enabling operators to quickly prototype friendly and opposing force behaviors that can learn and adapt. This would not only allow both civilian and military personnel to create realistic, operation-specific CGF behaviors in a timely manner, but also would eliminate the long delays and procurement costs normally associated with CGF development/modification by programmers. KATrix's NeuRule Intelligent Agent Technology meets this need by allowing "smart opponent" and "virtual teammate" CGFs to be created that can continuously learn from their interactions with manned forces, human instructors, and their own mistakes. The Phase I results demonstrate the effectiveness of the NeuRule Intelligent Agent learning in CGF applications. Phase II will extend these results by enabling SAF operators to develop, through use of an appropriate graphical user interface, CGF behaviors specific to dismounted-infantry operations in an Urban Environment. One of the main goals of the Phase II effort will be to demonstrate that SAF operators can gradually shift thier attention from low-level control tasks to high-level coordinated plans of action as smart opponent and virtual teammate CGFs gain competence learning relevant urban assault tactics and evasive manuevers.

Benefits:
Successful completion of the project will lead to the maturation of an innovative control technology enabling intelligent agents to learn from both supportive and adversarial sources. This technology is applicable to Army SAF Systems (CCTT), aviation SAF systems (AVCATT, WISSARD, TACTS range, etc.), undersea training, and weapons development (War breaker), as well as commercial interactive entertainment systems.