It is my belief that robotics is approaching a time when it can turn the corner from prohibitively expensive commercial systems to a technology that will be widely available. Computers have come through this transition over the past twenty years or so, but robots are moving more slowly through this curve, due in part to the higher cost and slower improvement of the mechanical components crucial to sensing and effecting the world. I believe current AI robotics research may move us more quickly through this transition than will the tradition robotics approaches of precise control, complex kinematics, etc.
We are now coming to a point when the software capabilities available on lower cost robots are reaching a point where they can overcome many of the limitations of the physical components -- the embedded processors can use new approaches to overcome the weaknesses of their cheaper components. To make this come about, we must start examining our robotics work with an eye towards this picture of more complex software architectures being used to control less precise devices. Precision controllers require expensive sensors and special purpose devices, but behavior-based robotics work has shown that architectural innovation can result in competing capabilities at significantly lower cost.
Work at the Autonomous Mobile Robotics Laboratory at the University of Maryland has been examining approaches to extending and understanding behavior-based robotics. Our work has included formalizing behaviors in terms of kinetic state machines and hybrid controllers and developing architectures for more complex behavior-based control. We are also examining human-in-the-loop (semi-autonomous) approaches where human reasoning can also be brought to bear without requiring dedicated control. In addition, we are now working towards the integration of situated planners and behavior-based robotics to allow goal-based reasoning to be realized using behavior-based components.