Push-Placement: A Hybrid Approach Integrating Prehensile and Non-Prehensile Manipulation for Object Rearrangement

TL;DR

Findings indicate that hybrid prehensile/non-prehensile action primitives can substantially improve efficiency in long-horizon rearrangement tasks.

Abstract

Efficient tabletop rearrangement remains challenging due to collisions and the need for temporary buffering when target poses are obstructed. Prehensile pick-and-place provides precise control but often requires extra moves, whereas non-prehensile pushing can be more efficient but suffers from complex, imprecise dynamics. This paper proposes push-placement, a hybrid action primitive that uses the grasped object to displace obstructing items while being placed, thereby reducing explicit buffering. The method is integrated into a physics-in-the-loop Monte Carlo Tree Search (MCTS) planner and evaluated in the PyBullet simulator. Empirical results show push-placement reduces the manipulator travel cost by up to 11.12% versus a baseline MCTS planner and 8.56% versus dynamic stacking. These findings indicate that hybrid prehensile/non-prehensile action primitives can substantially improve efficiency in long-horizon rearrangement tasks.

Figures (7)

• Figure 1: Two-object swap: (a) traditional pick-and-place requires buffering; (b) push-placement resolves the swap with fewer actions by pushing obstructing object(s) while placing the held object.
• Figure 2: Top-down example of push-placement. Colored rectangles depict current object footprints; dashed-outline rectangles (no fill) indicate target footprints. The gripper is depicted as a small filled circle on the grasped target. The target is aligned with its goal and advanced from the selected side to sweep blocking objects out of the goal region before placement.
• Figure 3: Side view of push-placement. The gripper holds the target above the table and moves it along the chosen side to sweep blockers out of the goal region; the target is then placed at the goal pose once the region is clear.
• Figure 4: Physics-in-the-loop execution. At each iteration, a complete plan from the current state to the goal is computed; only its first action is executed. The updated state is then considered and the process repeats, accommodating contact-induced changes (e.g., rotations) before committing to subsequent actions.
• Figure 5: Unsafe chained push scenario. The grasped target would push a blocker that immediately contacts a second object, creating a contact chain that can induce large, unpredictable rotations and potentially cause a newly rotated object to occlude a previously clear goal region. Such pushes are rejected by the admissibility filter (no secondary contacts allowed). Unfilled dashed-outline rectangles indicate goal footprints.