To work effectively in a dynamic and uncertain environment, an automated manufacturing system needs to be able to carry out actions that achieve its objectives, despite the broad range of, possibly unpredictable, events that may occur. Even in such environments, there may be regularities in behavior. These regularities can be embodied as assumptions, and used to simplify the construction of a controller for the system. However, at any time, only a subset of all assumptions might hold, and a controller relying on an assumption that does not hold will incurr errors. Therefore, it is necessary to continuously monitor assumptions and to dynamically adapt the controller when the set of working assumptions change. This paper describes an approach, called the Planner- Reactor approach, that exploits knowledge of regularities of the environment to incrementally build, and then continually adapt, a controller that operates effectively in the current vironment. The theory of the planner-reactor approach is briefly described, but the focus of the paper is on experimental results from an implementation of an industrial kitting robot using the approach.