An Expressive Language and Efficient Execution System for Software Agents

摘要

Software agents can be used to automate many of the tedious, time-consuminginformation processing tasks that humans currently have to complete manually.However, to do so, agent plans must be capable of representing the myriad ofactions and control flows required to perform those tasks. In addition, sincethese tasks can require integrating multiple sources of remote information ?typically, a slow, I/O-bound process ? it is desirable to make execution asefficient as possible. To address both of these needs, we present a flexiblesoftware agent plan language and a highly parallel execution system that enablethe efficient execution of expressive agent plans. The plan language allowscomplex tasks to be more easily expressed by providing a variety of operatorsfor flexibly processing the data as well as supporting subplans (formodularity) and recursion (for indeterminate looping). The executor is based ona streaming dataflow model of execution to maximize the amount of operator anddata parallelism possible at runtime. We have implemented both the language andexecutor in a system called THESEUS. Our results from testing THESEUS show thatstreaming dataflow execution can yield significant speedups over bothtraditional serial (von Neumann) as well as non-streaming dataflow-styleexecution that existing software and robot agent execution systems currentlysupport. In addition, we show how plans written in the language we present canrepresent certain types of subtasks that cannot be accomplished using thelanguages supported by network query engines. Finally, we demonstrate that theincreased expressivity of our plan language does not hamper performance;specifically, we show how data can be integrated from multiple remote sourcesjust as efficiently using our architecture as is possible with astate-of-the-art streaming-dataflow network query engine.