Robotic Manipulation of Objects Through Dual-Arm Handover Planning

Object manipulation without relying on complex fixtures remains a largely unresolved issue in industrial robotics, being generally limited to pick-and-place operations of easy to grasp objects. This work presents an adaptable manipulation planning algorithm for dual-arm robots, aiming to reorient an object from an initial position to a specified goal configuration without the need of external fixtures. Our approach integrates a precomputed regrasp graph with an online optimal handover planner that transforms the high-level sequence searched from the graph into executable grasp and handover poses. This approach reduces the overall graph complexity and enhances planning efficiency by merging high-level optimal sequence planning with the execution of predetermined motion primitives. The proposed algorithm is validated using different types of objects and a collaborative dual-arm robot. We also propose a comparison between our method and two benchmark approaches on a bin picking use case, to demonstrate how our pipeline improves the task efficiency.