Self-Assurance Modules for Autonomous Systems (SAMAS)

Self-Assurance Modules for Autonomous Systems (SAMAS) is designed to “bolt-on” to existing systems to incrementally increase assurance while in use. SAMAS leverages the current approach of Run Time Assurance, adding Run Time Validation and Run Time Argument Generation. While Run Time Assurance protects against known bad behaviors, Run Time Validation records and collates known good behaviors. Run Time Argument Generation seeks to “understand and explain” behaviors from conditions that are neither bad or good. Each time an autonomous system performs as expected, it is recorded in a graph, allowing certifiers to trust that the system behaves repeatedly in tested conditions. In untested conditions, SAMAS generates a model of the environmental conditions and the system’s response. SAMAS then generates an assurance case fragment to explain the condition experienced and the action taken. Over time SAMAS increases transparency of learning systems even after deployment, allowing increased capability in dynamic and uncertain environments. SAMAS leverages a unique technique to encode dynamic reachability calculations into a neural fuzzy network. This allows compression of computationally expensive trajectory calculations. Leveraging this technology and the ability to translate fuzzy networks into explainable rules, we can describe the actions of an autonomous control system in previously unobserved conditions.