The reasons for integrating collision avoiding path planning into a task-level programmable multi-sensor robot system are put forward. The underlying system architecture and the specific approaches for environment modelling, task planning and path planning are discussed. Task planning is performed using a rule based expert sytem and a frame representation of relevant environment data. Path planning is based on a configuration-space approach with a fast new algorithm for obstacle transformation. Results gained from experimental laboratory work are presented and show some advantages and problems of the entire system.