Efficient Integration of Task and Motion Planner for Manipulation Tasks Using Geometric Spanner Graph

Integrating task and motion planning is challenging as task planners work in the discrete domain and computationally expensive motion planners in the continuous domain. A single-task action can be translated to infinite motion planning instances, resulting in a trade-off between computational cost and completeness. This paper proposes a novel approach to improve the performance of integrated task and motion planners by exploiting the spatial relationship of the objects, using a geometric spanner graph, namely, Yao graph. The proposed interface utilizes the spatial relationship to build constraints for the task planner, sample geometrically feasible poses for the motion planner to achieve and find objects to remove on failed motion planning attempts. The results demonstrate using spatial relationship improves the total planning time and success rate in benchmark environments for manipulation planning compared to traditional task and motion planning. The proposed method can automatically detect the inter-dependencies among movable objects with minimal overhead computation.