TAPOM: Task-Space Topology-Guided Motion Planning for Manipulating Elongated Object in Cluttered Environments

TL;DR

TAPOM addresses the challenge of manipulating elongated objects in cluttered, narrow environments by decoupling a topology-driven, task-space analysis from the high-dimensional configuration-space search. It builds a topology graph of obstacle contacts $G_{O_{ ext{seg}}}$ to extract channels $ C$ via simple loops, then abstracts these into a channel graph $G_{ ext{ch}}$ with node and edge weights to score high-level paths $\mathcal{P}^*$ using the composite score $W(\mathcal{P}_k)$. The low-level planner then samples keyframes within the selected channels and uses a sequence of RRT variants, including budgeted BiRRT, to fuse these keyframes into a collision-free trajectory $\mathcal{J}$ from $q_{ ext{start}}$ to $q_{ ext{goal}}$. Experiments across five cluttered scenarios and real-world robot tests demonstrate TAPOM’s superior success rates and shorter planning times relative to baselines, with ablations confirming the essential roles of topology-aware planning and keyframe prioritization. The work offers a scalable, structure-aware approach for complex manipulation tasks and lays groundwork for future integration with optimization-based smoothing and perception-driven topology inference.

Abstract

Robotic manipulation in complex, constrained spaces is vital for widespread applications but challenging, particularly when navigating narrow passages with elongated objects. Existing planning methods often fail in these low-clearance scenarios due to the sampling difficulties or the local minima. This work proposes Topology-Aware Planning for Object Manipulation (TAPOM), which explicitly incorporates task-space topological analysis to enable efficient planning. TAPOM uses a high-level analysis to identify critical pathways and generate guiding keyframes, which are utilized in a low-level planner to find feasible configuration space trajectories. Experimental validation demonstrates significantly high success rates and improved efficiency over state-of-the-art methods on low-clearance manipulation tasks. This approach offers broad implications for enhancing manipulation capabilities of robots in complex real-world environments.

Figures (7)

• Figure 1: A typical manipulation task. A robot needs to manipulate an elongated object (a rebar beam) through a narrow passage (free sapces in a scaffold). Because translational motion is restricted, this robot needs to align object with the passage to pass through.
• Figure 2: Overview of the proposed planner. (a) Obstacles are manually segmented into some sub-obstacles like boxes, cylinders, and spheres. Contact points between sub-obstacles are identified as red dots. (b) Connectivity graph $\mathcal{O}_{\text{seg}}$ is represented by nodes (red dots) and edges (black line). Simple loops (shaded areas, light red indicates invalid loops) are detected in this graph, representing potential channels. Arrows are several candidate channel paths with different feasibility. (c) Channel extraction: plane $\Pi$ is fitted to loop contact points $\{p_1, \dots, p_m\}$ via least-squares. Channel area (white) derived from convex hull on $\Pi$, thickness from perpendicular clearance. (d) Edge weights in channel graph $G_{\text{ch}}$: edge weights $w_{e_{ij}}$ indicate transition feasibility. Additionally, optimal high-level path $\mathcal{P}^*$ is highlighted in red. (e) Channel connectivity graph $G_{\text{ch}}$: nodes (blue dots) represent channels and edges represent feasible transitions between them. Channel paths in (b) are generated in this graph. (f) Generation of keyframes from the optimal high-level path $\mathcal{P}^*$ in task space. (g) Growing and merging trees within keyframe regions.
• Figure 3: Performance of TAPOM in different scenarios. Light yellow area indicates the swept volume of robots and the grasped object, which starts from initial configuration and ends at goal configuration. (a) Part delivery through industrial shelving (Part Delivery, PD). (b) Installing a ceiling tile in a civil infrastructure (Ceiling Installation, CI), defined in liang2022trajectory. (c) Robotic insertion of a rebar into a rebar cage typically used for rebar-reinforced concrete structures, inorder to weld a rebar to a skeleton (Rebar Assembly, RA), defined in momeni2022automated. (d) Delivering a jack to a trapped man and helping him support potentially falling rocks (Rescue). Red areas indicate free space between rocks that enables robots to move. (e) Beam transportation in an unfinished building (Beam Transportation, BT).
• Figure 4: Results of experiments on different scenarios.
• Figure 5: Results of real-world experiments. (a) The configurations of the real-world experiment. (b) The process of manipulation passing through a shelf full of devices and daily tools.