Abstract: | This paper gives an overview of the main current research project in> assembly robotics at the Department of Artificial Intelligence, Edinburgh, which began in 1986. Most research work in assembly robotics, including our own previous work in this Department, follows from a particular attitude to the problems of assembly which we now call the ``classical approach'' or the ``Newtonian approach'' to distinguish it from our new approach, which we call the ``behaviour-based approach''. We think that some of the difficult problems of current methods of programming assembly robots are artefacts of the Newtonian approach. These problems become particularly evident when trying to generate assembly programs off-line from some general specification of the task, and when trying to use sensors to handle uncertainty of location and dimension. The main feature of the approach is the use of a hybrid system in which a classical knowledge-based planner working in an idealised world makes plans to be interpreted by a behaviour-based assembly agent. This agent is designed with a Brooksian behaviour-based philosophy, but with the very unBrooksian feature that it is designed to be an interpreter of plans, and designed to simplify the planner as much as possible.
|