MO-SeGMan: Rearrangement Planning Framework for Multi Objective Sequential and Guided Manipulation in Constrained Environments

TL;DR

MO-SeGMan tackles constrained rearrangement planning by jointly optimizing an object placement sequence and the corresponding motion plans. It introduces Selective Guided Forward Search (SGFS) to relocate only critical obstacles and employs a lazy evaluation strategy to refine the sequence using $K$-object interactions and RRT-based costs. A subgoal refinement step further reduces unnecessary pick-and-place actions, improving solution quality without sacrificing feasibility. Extensive experiments across 14 tasks demonstrate robust performance, faster solution times, and superior quality compared to baselines, highlighting the method's scalability to cluttered, real-world environments. The framework holds promise for warehouse and household automation, with potential extensions to multi-robot collaboration.

Abstract

In this work, we introduce MO-SeGMan, a Multi-Objective Sequential and Guided Manipulation planner for highly constrained rearrangement problems. MO-SeGMan generates object placement sequences that minimize both replanning per object and robot travel distance while preserving critical dependency structures with a lazy evaluation method. To address highly cluttered, non-monotone scenarios, we propose a Selective Guided Forward Search (SGFS) that efficiently relocates only critical obstacles and to feasible relocation points. Furthermore, we adopt a refinement method for adaptive subgoal selection to eliminate unnecessary pick-and-place actions, thereby improving overall solution quality. Extensive evaluations on nine benchmark rearrangement tasks demonstrate that MO-SeGMan generates feasible motion plans in all cases, consistently achieving faster solution times and superior solution quality compared to the baselines. These results highlight the robustness and scalability of the proposed framework for complex rearrangement planning problems.

Figures (8)

• Figure 1: Rearrangement planning cases of varying difficulty. The objective is to generate a motion plan for the robot (yellow) to move the goal objects (colored) to their designated goal locations (silhouettes), while interacting with movable obstacles (white) and other goal objects.
• Figure 2: Steps in object placement sequence generation: (a) collision checking along the object placement trajectories $\mu_{o_i}^{\text{place}}$, (b) the dependency graph $G$ (dotted edges: weak dependencies, solid edges: strong dependencies), (c) acyclic dependency graph $G'$, and (d) optimized object placement sequence $\mathcal{S}$.
• Figure 3: Flowchart of the motion planning process. The pick and placement plan are decomposed due to the complex nature of the problem. If a motion plan cannot be generated, the movable obstacles are relocated with SGFS.
• Figure 4: The Selective Guided Forward Search (SGFS) explores configurations for the relocation of critical objects to obtain a feasible task trajectory.
• Figure 5: Relocation-point generation: (a) Local Occupancy Matrix (LOM) with object mask (red box), (b) convolution result (light areas), (c) Euclidean Distance Transform (green = higher clearance), (d) candidate relocation points (light areas).