A Framework for Intelligent Autonomy

摘要

This paper describes the design and implementation of Draper Laboratory’s All-rnDomain Execution and Planning Technology (ADEPT) architecture for intelligentrnautonomy. Intelligent autonomy is the ability to plan and execute complex activities in arnmanner that provides rapid, effective response to stochastic and dynamic mission events.rnThus, intelligent autonomy enables the high-level reasoning and adaptive behavior for anrnunmanned vehicle that is provided by an operator in man-in-the-loop systems.rnDraper’s intelligent autonomy has architecture evolved over a decade and a halfrnbeginning in the mid 1980’s [6, 8, 9 and 10] culminating in an operational experimentrnfunded under DARPA’s Autonomous Minehunting and Mapping Technologies (AMMT)rnunmanned undersea vehicle program [1]. ADEPT continues to be refined through itsrnapplication to current programs that involve air vehicles and ground robots, with thernobjective of solidifying a proven, dependable software approach that can be quicklyrnapplied to new vehicles and domains.rnThe architecture can be viewed as a hierarchical extension of the sense-think-actrnparadigm of intelligence and has strong parallels with the military’s Observe-Orient-rnDecide-Act (OODA) loop [2]. The key elements of the architecture are planning andrndecision-making nodes comprising modules for situation assessment, plan generation,rnplan implementation and coordination. A reusable, object-oriented software frameworkrnhas been developed that implements these functions. As the architecture is applied tornnew areas, only the application specific software needs to be developed.rnThis paper describes the core architecture in detail and discusses how this has beenrnapplied in the undersea, air, ground and space domains.