Hierarchical PDDL Task Planning and Transfer for Home Service Robots

This paper addresses the challenge of complex task planning for home service robots by integrating semantic parsing and the Planning Domain Definition Language (PDDL). We present a task planner to enable robots to comprehend user commands, automatically generate problem files, and plan sequences of atomic actions. Current household robots are limited in their ability to handle complex user-defined tasks due to restrictions in task transferability and real-time performance. To overcome these limitations, we introduce a hierarchical approach to task planning, which refines tasks and objects from abstract to concrete levels, significantly reducing computational complexity and improving adaptability across different environments. We validate our framework through experimental demonstrations in a simulated kitchen environment using Unity3D, where it performs complex tasks such as preparing afternoon tea. Our results show a significant reduction in planning time compared to traditional flat task planners, making real-time task execution feasible. Furthermore, the hierarchical structure enhances the transferability of task knowledge to new environments, improving the robot's ability to adapt to different home environments.