LaGarNet: Goal-Conditioned Recurrent State-Space Models for Pick-and-Place Garment Flattening

2025/08/23 by Kadi, Halid Abdulrahim, Terzić, Kasim
#FOS: Computer and information sciences #Robotics (cs.RO)

paper · doi:10.48550/arxiv.2508.17070

Abstract

We present a novel goal-conditioned recurrent state space (GC-RSSM) model capable of learning latent dynamics of pick-and-place garment manipulation. Our proposed method LaGarNet matches the state-of-the-art performance of mesh-based methods, marking the first successful application of state-space models on complex garments. LaGarNet trains on a coverage-alignment reward and a dataset collected through a general procedure supported by a random policy and a diffusion policy learned from few human demonstrations; it substantially reduces the inductive biases introduced in the previous similar methods. We demonstrate that a single-policy LaGarNet achieves flattening on four different types of garments in both real-world and simulation settings.

Citations

• Diffusion Dynamics Models with Generative State Estimation for Cloth Manipulation
• Learning World Models for Unconstrained Goal Navigation
• DRAPER: Towards a Robust Robot Deployment and Reliable Evaluation for Quasi-Static Pick-and-Place Cloth-Shaping Neural Controllers
• SIS: Seam-Informed Strategy for T-shirt Unfolding
• UniGarmentManip: A Unified Framework for Category-Level Garment Manipulation via Dense Visual Correspondence
• Learning Keypoints for Robotic Cloth Manipulation using Synthetic Data
• Facing Off World Model Backbones: RNNs, Transformers, and S4
• FabricFolding: Learning Efficient Fabric Folding without Expert Demonstrations
• Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
• Diffusion policy: Visuomotor policy learning via action diffusion
• PlaNet-ClothPick: Effective Fabric Flattening Based on Latent Dynamic Planning
• Mastering Diverse Domains through World Models
• Learning Fabric Manipulation in the Real World with Human Videos
• Cloth Funnels: Canonicalized-Alignment for Multi-Purpose Garment Manipulation
• SpeedFolding: Learning Efficient Bimanual Folding of Garments
• Mesh-based Dynamics with Occlusion Reasoning for Cloth Manipulation
• FabricFlowNet: Bimanual Cloth Manipulation with a Flow-based Policy
• Learning Visible Connectivity Dynamics for Cloth Smoothing
• LazyDAgger: Reducing Context Switching in Interactive Imitation Learning
• Cloth Manipulation Planning on Basis of Mesh Representations with Incomplete Domain Knowledge and Voxel-to-Mesh Estimation
• VisuoSpatial Foresight for Physical Sequential Fabric Manipulation
• SoftGym: Benchmarking Deep Reinforcement Learning for Deformable Object Manipulation
• Learning Arbitrary-Goal Fabric Folding with One Hour of Real Robot Experience
• Long-Horizon Visual Planning with Goal-Conditioned Hierarchical Predictors
• VisuoSpatial Foresight for Multi-Step, Multi-Task Fabric Manipulation
• Learning Predictive Representations for Deformable Objects Using Contrastive Estimation
• CLOTH3D: Clothed 3D Humans
• Dream to Control: Learning Behaviors by Latent Imagination
• Learning to Manipulate Deformable Objects without Demonstrations
• Visual Foresight: Model-Based Deep Reinforcement Learning for Vision-Based Robotic Control
• Learning Latent Dynamics for Planning from Pixels
• Sim-to-Real Reinforcement Learning for Deformable Object Manipulation
• World Models
• Hindsight Experience Replay
• Image quality assessment: from error visibility to structural similarity

Related

• PlaNet-ClothPick: Effective Fabric Flattening Based on Latent Dynamic Planning
• GraphGarment: Learning Garment Dynamics for Bimanual Cloth Manipulation Tasks
• Beyond Static Perception: Integrating Temporal Context into VLMs for Cloth Folding
• Recognising Known Configurations of Garments For Dual-Arm Robotic Flattening
• Deep Imitation Learning of Sequential Fabric Smoothing From an Algorithmic Supervisor