ConTrAkt: Constraint-Based Online Transformation of Abstract Plans into Executable Robot Actions

Monday, Apr 3, 2017

In recent years, research in the field of autonomous mobile robots operating in human-populated environments has seen significant progress. However, when it comes to robustly executing high-level robot plans, problems remain. The reason is that the execution of seemingly simple actions such as driving to a specific location requires many interdependent system components working together in a complex way; this yields a plethora of potential error sources.

While approaches exist that try to solve this coordination problem, there is no explicit separation between the task of the robot and the system requirements from a user point of view. The disadvantages are that the planning task becomes more complex and applications are specific to a particular platform. This in turn requires deep system understanding on the side of the high-level agent designer.

In contrast, this project envisions to separate the high-level task from internal requirements of how to execute it on the robot. The idea is to transform a particular plan at runtime in such a way that the required system components together with their dependencies are automatically integrated based on a declarative language specification. At the same time, information needed for execution monitoring and and possible error recovery will be provided. The results will be evaluated on different mobile robots in domestic environments.

Funding Agency

This project is funded by the German Research Foundation DFG with project number 288705857 (GL-747/23-1).

This project is in cooperation with the MASCOR Institute, FH Aachen University of Applied Sciences.