Towards Autonomous Systems: From Control Systems to Intelligent Control to Intelligent Behavior Generation to Cooperative Autonomy

Today a number of enabling technologies have matured to the point that it is possible to build robots and intelligent machines that are physically capable of autonomous behavior. However, to achieve the promise of autonomy, we also need equivalently-mature information-processing and decision systems to exploit these physical capabilities. In this talk we discuss the problem of devising truly autonomous systems in three parts, describing several threads of research from the speaker's experience. In Part 1, we begin with a discussion of intelligent control, including its promise and reality, introducing specifically the paradigm of iterative learning control (ILC). After highlighting and critiquing the history and accomplishments of ILC, we posit that in fact intelligent control has not achieved its promise and argue that as we try to develop increasingly autonomous systems we need better understandings of the purpose (goals), the components (memory, learning), and organization of intelligence (models, language, architecture). From this motivator, in Part 2 we consider how to move beyond conventional intelligent control to develop intelligent behavior generators for single-system autonomy, focusing on mobile robots operating in semi-structured environments. We present an intelligent, reactive command and control system that uses a multi-resolution, hierarchical task-decomposition strategy based on a grammar of atomic actions. The effectiveness of the strategy is demonstrated in actual tests with real robots in which the path-planning and control algorithms are implemented in a distributed processing environment.