[論文レビュー] ReloPush-BOSS: Optimization-guided Nonmonotone Rearrangement Planning for a Car-like Robot Pusher

Jeeho Ahn, Christoforos Mavrogiannis|arXiv (Cornell University)|Jan 29, 2026

Robot Manipulation and Learning被引用数 0

ひとこと要約

ReloPush-BOSS は最適化誘導Prerelocationと push-traversability グラフおよび深さ優先探索を統合し、密集クラスター下の非monotone 多目的再配置タスクを解決します。車型プッシャーを用いた実機検証でロバスト性を示します。

ABSTRACT

We focus on multi-object rearrangement planning in densely cluttered environments using a car-like robot pusher. The combination of kinematic, geometric and physics constraints underlying this domain results in challenging nonmonotone problem instances which demand breaking each manipulation action into multiple parts to achieve a desired object rearrangement. Prior work tackles such instances by planning prerelocations, temporary object displacements that enable constraint satisfaction, but deciding where to prerelocate remains difficult due to local minima leading to infeasible or high-cost paths. Our key insight is that these minima can be avoided by steering a prerelocation optimization toward low-cost regions informed by Dubins path classification. These optimized prerelocations are integrated into an object traversability graph that encodes kinematic, geometric, and pushing constraints. Searching this graph in a depth-first fashion results in efficient, feasible rearrangement sequences. Across a series of densely cluttered scenarios with up to 13 objects, our framework, ReloPush-BOSS, exhibits consistently highest success rates and shortest pushing paths compared to state-of-the-art baselines. Hardware experiments on a 1/10 car-like pusher demonstrate the robustness of our approach. Code and footage from our experiments can be found at: https://fluentrobotics.com/relopushboss.

研究の動機と目的

• Address nonmonotone multi-object rearrangement in densely cluttered spaces using a car-like pusher.
• Integrate kinematic, geometric, and physics constraints via a unified object traversability graph.
• Improve prerelocation decisions with optimization informed by Dubins path classification.
• Mitigate local minima through seeded warm-starts and backtracking search.
• Demonstrate scalability up to 13 objects and validate on hardware.

提案手法

• Construct a Push-Traversability (PT) graph with pushing poses on object boundaries and feasible Dubins paths between poses.
• Embed prerelocations as intermediate Dubins paths to connect push segments when direct transfer is infeasible.
• Formulate prerelocation optimization to minimize the sum of lengths of the two push segments surrounding the prerelocation.
• Use seeded warm-starts derived from Dubins-path-based seeds to avoid high-cost local minima during optimization.
• Employ a depth-first search with backtracking over rearrangement sequences to manage combinatorial complexity.
• Operate within a constrained SE(2) workspace with nonholonomic car-like kinematics and quasistatic pushing assumptions.
• Guarantee conditional completeness within the defined primitives (at most one prerelocation per object and straight-line obstacle clearing).

実験結果

リサーチクエスチョン

• RQ1 How can prerelocation decisions be optimized to reduce total push-transfer distance in dense clutter?
• RQ2 Can Dubins-path classification inform robust initialization to avoid poor local minima in prerelocation optimization?
• RQ3 Does incorporating optimization-guided prerelocations into a traversability graph improve success rates and path lengths for car-like pushers in nonmonotone rearrangements?
• RQ4 What is the impact of backtracking high-level planning on scalability to larger object sets?
• RQ5 How well does the approach transfer to real hardware beyond simulation?

主な発見

• ReloPush-BOSS achieves higher success rates across dense scenarios with up to 13 objects than ablated versions and literature baselines.
• Optimized prerelocations reduce total push-transfer length and overall path length compared to non-optimized prerelocations.
• Seeded warm-starts based on Dubins path insights effectively avoid high-cost local minima during prerelocation optimization.
• A depth-first search with backtracking improves robustness and scalability over purely greedy or exhaustive approaches.
• Hardware demonstrations on a 1/10-scale car-like pusher validate robustness in real-world friction and tracking conditions.