← Back ICRA 2026

Supportive Relationships-Aware Hierarchical Reinforcement Learning for Efficient Ex-Situ Object Rearrangement

leibing xiao, Xuemei Wang, Zhao Zhongqiang, Yachao Wang, Jin Liu, Chaoqun Wang

PDF

AI summary

Key figure (auto-extracted from paper)

Integrating LLM-derived supportive relationships into hierarchical reinforcement learning cuts ex-situ object rearrangement time by roughly half.

Object rearrangement Hierarchical reinforcement learning Supportive relationships Graph capsule networks Large language models Robotic stacking

Problem

Robots inefficiently handle multiple objects individually in open environments, causing high movement costs and long completion times during ex-situ rearrangement tasks.

Approach

An LLM maps object support and semantic links, which guide a hierarchical reinforcement learning framework to intelligently group compatible items for stacking via high-level attention and low-level graph capsule networks.

Key results

Reduces task completion time by approximately 50% compared to non-supportive baselines
Lowers the number of required stacking groups and robot movements
Outperforms ACO and CapAM baselines across 30, 40, and 50 object scenarios
Successfully validated in both simulation and real-world robotic experiments

Why it matters

Enables scalable, efficient multi-object manipulation for service and industrial robots operating in complex, open-world settings.

Abstract

In ex-situ object rearrangement tasks within open environments, robots face significant challenges due to the increased cost of moving objects over large workspaces. To address this issue, we propose a hierarchical reinforcement learning-based approach that takes into account the supportive relationships and semantic correlations between objects. The robot groups and stacks objects with compatible supportive capabilities, moving them together to their target locations to optimize task execution. Specifically, we use a large language model to assess the supportive relationships and semantic correlations between objects. In the high-level decision-making process, objects are grouped based on their supportive capabil- ities, while the low-level process refines these groupings using a graph capsule convolutional network. Experimental results demonstrate that our approach not only reduces the number of movements required but also improves task efficiency and significantly decreases task completion time by approximately 50%, compared to methods that do not consider supportive relationships.

Index terms

Domestic Robotics Service Robotics Planning Scheduling and Coordination