Simulation Modeling for Safe AI (SiMSA)

Central to SiMSA is RTSync's Automatic Scenario Construction/Selection (ASC) technology that creates a capability to model and simulate complex System of Systems (SoS) by composing them from component system models in various combinations and configurations. Sponsored by the Missile Defense Agency (MDA), ASC aggregates models of elements, environments, threats, and defenses from repositories into simulation executable form. We are currently demonstrating ASC and SiMSA capabilities with advanced synthetic aperture radar deep learning applications sponsored by the Air Force Research Lab (AFRL).

The SiMSA software toolset provides effective safety-focused training methods for AI and ML-based systems which are essential for achieving safe performance of such systems in realistic environments. SiMSA is capable of generating voluminous scenarios portfolios that generate labeled data for training that stretch the limits of systems beyond the relatively narrow spaces for which they have been designed. Alternative approaches employ virtual reality-based immersive environments for training designed systems. Such systems are very difficult and costly to develop with the high fidelity required.