Regression-Based Task Dependency Decomposition for Mobile Manipulation with Symbolic Planning

Symbolic planning algorithms solve task planning problems by generating a sequence of symbolic actions for a single robot. Borrowing the idea of regression planning, we introduce Regression-based Task Dependency Decomposition (RTDD) that extracts the structural relationships within symbolic planning solutions generated by symbolic planners. This enables the decomposition of mobile manipulation problems into a sequence of ordered intermediate goals. By leveraging a reliable symbolic planner which can be treat as a black box, our approach breaks down complex tasks into critical intermediate goals while establishing a partial order among them. The outputs of our method are both interpretable and robust, facilitating efficient task allocation in multi-robot systems to achieve overarching goal.