Geometry-aware RL for Manipulation of Varying Shapes and Deformable Objects

2025-02-10

AI Summary

This paper addresses the challenge of modeling and controlling dexterous robotic manipulation of multi-modal objects—such as rigid bodies, ropes, and cloth—by proposing HEPi, the first SE(3)-equivariant heterogeneous geometric graph framework. HEPi unifies the 3D interaction modeling among actuators, objects, and environment, explicitly encoding functional heterogeneity and 3D geometric symmetries, while supporting multi-actuator coordination and zero-shot generalization to unseen object geometries. Technically, it integrates heterogeneous graph neural networks, SE(3)-equivariant message passing, and reinforcement learning. Evaluated on novel benchmarks—including rigid-body insertion, rope threading, and cloth hanging—HEPi achieves substantial improvements: +28.6% in average return, 3.2× higher sample efficiency, and superior cross-shape generalization, consistently outperforming both Transformer-based and homogeneous equivariant baselines.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Computer Vision: Multi-modal Vision

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Manipulating objects with varying geometries and deformable objects is a major challenge in robotics. Tasks such as insertion with different objects or cloth hanging require precise control and effective modelling of complex dynamics. In this work, we frame this problem through the lens of a heterogeneous graph that comprises smaller sub-graphs, such as actuators and objects, accompanied by different edge types describing their interactions. This graph representation serves as a unified structure for both rigid and deformable objects tasks, and can be extended further to tasks comprising multiple actuators. To evaluate this setup, we present a novel and challenging reinforcement learning benchmark, including rigid insertion of diverse objects, as well as rope and cloth manipulation with multiple end-effectors. These tasks present a large search space, as both the initial and target configurations are uniformly sampled in 3D space. To address this issue, we propose a novel graph-based policy model, dubbed Heterogeneous Equivariant Policy (HEPi), utilizing $SE(3)$ equivariant message passing networks as the main backbone to exploit the geometric symmetry. In addition, by modeling explicit heterogeneity, HEPi can outperform Transformer-based and non-heterogeneous equivariant policies in terms of average returns, sample efficiency, and generalization to unseen objects.

Problem

• Geometry-aware RL for manipulation
• Handling varying shapes and deformable objects
• Graph-based policy model for precise control

Innovation

• Heterogeneous graph representation
• SE(3) equivariant message passing
• Heterogeneous Equivariant Policy (HEPi)