ALORE

Autonomous Large-Object Rearrangement
with a Legged-Manipulator

arXiv Code

Logo ALORE : Autonomous Large-Object
Rearrangement with a Legged-Manipulator

arXiv Code

Abstract

Endowing robots with the ability to rearrange various large and heavy objects, such as furniture, can substantially alleviate human workload. However, this task is extremely challenging due to the need to interact with diverse objects and efficiently rearrange multiple objects in complex environments while ensuring collision-free loco-manipulation. In this work, we present ALORE, an Autonomous Large-Object REarrangement system for a legged manipulator that can rearrange various large objects across diverse scenarios. The proposed system is characterized by three main features: (i) a hierarchical reinforcement learning training pipeline for multi-object environment learning, where a high-level object velocity controller is trained on top of a low-level whole-body controller to achieve efficient and stable joint learning across multiple objects; (ii) two key modules, a unified interaction configuration representation and an object velocity estimator, that allow a single policy to regulate planar velocity of diverse objects accurately; and (iii) a task-and-motion planning framework that jointly optimizes object visitation order and object-to-target assignment, improving task efficiency while enabling online replanning. Comparisons against strong baselines show consistent superiority in policy generalization, object-velocity tracking accuracy, and multi-object rearrangement efficiency. Key modules are systematically evaluated, and extensive simulations and real-world experiments are conducted to validate the robustness and effectiveness of the entire system, which successfully completes 8 continuous loops to rearrange 32 chairs over nearly 40 minutes without a single failure, and executes long-distance autonomous rearrangement over an approximately 40m route.

Overview

1. Method

(a) Workflow of the proposed system, including tailored perception, robot-object system planning, a hierarchical RL-based controller, and object grasp-and-release components. (b) Training pipeline of the hierarchical controller, which consists of the low-level WBC and the high-level object controller. After training, the learned policy can be directly deployed on a legged manipulator without further fine-tuning.

In this work, our goal is to autonomously rearrange various kinds of large objects, such as chairs, tables, and bins, to their target positions and orientations using a legged manipulator. To this end, a complete large-object rearrangement system is shown in (a), which mainly consists of tailored perception, TAMP, low-level WBC, and high-level object velocity controller. First, given objects' target poses, the task planning algorithm generates task sequences (S1 → S2 → · · · → Si), including the object visitation order and the object-to-target assignment. Next, the coarse-to-fine robot trajectories for each subtask Si are computed based on the estimated object pose, which are then tracked by the low-level WBC. Once the robot reaches the target pose, it grasps the object, and a collision-free trajectory is planned for the object. This trajectory is tracked by the high-level object controller until the object is released at its goal pose. The whole process above is repeated until all subtasks are finished. In this task, the perception module provides global poses of the robot and objects. Similar to prior large-object rearrangement settings, we assume that the approximate positions of target objects are available (without requiring precise poses) and that feasible grasp regions are provided. It allows us to focus on the core challenges, i.e., robust post-grasp interaction control across diverse large objects.

2. Simulation

2.1 Comparison with Direct Method


2.2 Comparison of Task Planning with Greedy Method


2.3 Various Object Rearrangement in Diverse Scenarios

Library

Office
Warehouse

3. Real-World Experiments

3.1 Various Objects Rearrangement

Top-down View
Side View

3.2 Long-Horizon Multi-Chair Rearrangement


3.3 Long-Distance Object Rearrangement