Proc. SPIE. 10651, Open Architecture/Open Business Model Net-Centric Systems and Defense Transformation 2018
KEYWORDS: Complex systems, Computing systems, Telecommunications, System identification, Systems engineering, Data communications, Systems modeling, Information security, Computer security, Process engineering
Mission objectives for cyber systems operating in a tactical or deployed operational environment must address not only security but also resilience. A cyber resilient architecture is engineered for completing mission objectives in the “face of persistent, stealthy, and sophisticated attacks of cyber resources (MITRE, 2011)”. Ever evolving adversaries drive the need for system architectures to protect cyber resources but enable operations during an attack to achieve mission objectives. Similar to cyber security, resilience must be engineered into all layers of system architecture at inception, baking protections for security and redundancies for resilience through all layers of the system architecture.
A collaborative robotic team may need to allocate multiple tasks within an unknown and highly dynamic environment. Highly dynamic environments entail constantly changing states of its entities, objects, and situational characteristics. Tasks in dynamic environments can have unknown information, changing requirements, and can result in unforeseen goal states. (e.g., searching an unfamiliar building can result in an unknown number of rooms, or open areas to be further searched). Essentially, the task requirements can change, the asset/robot ability to perform the task can change, and the environment can change such that an allocation of tasks may need to be re-allocated. The allocation process must be flexible enough and ad-hoc in nature to compensate for such dynamics. This report presents the results of multiple investigations into various market-based, ad-hoc methods as a means to flexibly allocate tasks across a mobile robotic team in unknown and highly dynamic environments. These ad-hoc methods can be controlled from a centralized point or implemented in a decentralized mode. The decentralized control is of greater interest since it reduces processing bottlenecks, eliminates single points of failures, and can encourage network dataflow that is more natural to the ad-hoc nature of the mobile team of robotic vehicles. This is especially true when the number of tasks and robots are scaled up (i.e., swarm robotics). The approach utilizes weighted formulas that represent a robot’s ability to engage each of the identified tasks, and a task’s allocation is based on comparing the results of these weighted formulae. The allocation process is improved via optimizing the formulas’ weights based on deep-learning methods.