MIT Building 5-214 or 32-310
Cambridge, MA 02139
Phone: 617-253-1531 or 324-2613
Email: mikerb AT csail.mit.edu
|
|
Michael R. Benjamin MIT Building 5-214 or 32-310 Cambridge, MA 02139 Phone: 617-253-1531 or 324-2613 Email: mikerb AT csail.mit.edu |
|
Effective autonomous decision making in uncertain, dynamic
environments is a difficult problem. It is much more difficult when
there are multiple simultaneous objectives that are in competition or
confliction with each other. For a robot, the reason why this is hard
is that, with each concern or objective, there is a corresponding
"state" of the world, i.e., set of salient features of the environment
that impact the decision making. With multiple objectives comes a
potential explosion of state combinations, making a "global" mapping
of states to actions intractible.
This problem is important to research because it is both ubiquitous and because current approaches are limited. Typically, in robotic systems, objectives are either decoupled and addressed by strict priority, or objectives are addressed simultaneously by choosing decisions that reflect the "average" of separate good decisions. Both approaches are simple but flawed, and do not fair well compared to the performance of human decision making. Humans are very good at using common sense to exploit the flexibility and constraints inherent in the relationship between actions and goals, to ultimately achieve effective balance between objectives in their actions. My research has focussed on this problem by developing a novel new mathematical programming model, interval programming (IvP) and applying it to several areas in both autonomous robotic control, as well as in human decision aid systems. |