[en] Efficient use of industrial robots requires a strong interplay between user level commands, the motion control system, and external equipment. It should also be possible for an experienced application engineer to tailor the motion control to a specific application in a convenient way, instead of deficient utilization of the device or tricky user programming which is often the case today. A layered software architecture has been designed based on an application oriented view, considering typical hardware and software constraints. The top layers or the architecture support improved integration of off-line programming with interactive teach-in programming. The proposed solution is based on a transformation of robot programs between an on-line and an off-line representation. A central part of the architecture is an intermediate software layer, allowing the experienced user to introduce application specific motion primitives, on top of the motion control system. Flexibility during system configuration combined with computing efficiency and performance at run-time is of major importance. The solution is based on so called actions, which are methods to be passed between different software layers. Such methods can be specification of nonlinear control parameters, application specific control strategies, or treatment of external sensor signals. The actions can be implemented efficiently even in the multiprocessor case by using relocatable executable pieces of code generated from a special cross-compilation strategy. The lowest layers, comprising the motion control, have to be efficient and still fit in with the upper layers. In these layers, software solutions include an external sensor interface and a concept of motion pipelining allowing sensor based motions to be partly computed in advance. An experimental platform, built around commercially available robots, has been developed to verify the proposed solutions. (au)