An architecture for vision and action

摘要: Vision systems that have successfully supported nontrivial tasks invariably taken advantage of constraints derived from the task and environment to increase reliability lower complexity perception. We propose it is possible build a general purpose vision system, is, one can support wide variety tasks, take such constraints. The central idea within our proposed architecture reactive skill. Skills are concurrent control routines assembled at run time using instructions symbolic execution system. Visual modules used as resources in construction these skills. agent continuous feedback loops but constructed discrete, instructions. key general-purpose ability parametrize primitive elements system compose visual ways. demonstrate context an implemented example robot collecting trash off floor depositing garbage can.