ManiLong-Shot: Interaction-Aware One-Shot Imitation Learning for Long-Horizon Manipulation

2025/12/18 by Zixuan Chen, Chen, Zixuan, Chongkai Gao +9
Computer Science · Engineering · #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Reinforcement Learning in Robotics #Robot Manipulation and Learning #Robotics (cs.RO)

paper · pdf · doi:10.48550/arxiv.2512.16302

openalex publication_date 2025/12/18 · openalex created_date 2025/12/21 · openalex updated_date 2026/07/28

Abstract

One-shot imitation learning (OSIL) offers a promising way to teach robots new skills without large-scale data collection. However, current OSIL methods are primarily limited to short-horizon tasks, thus limiting their applicability to complex, long-horizon manipulations. To address this limitation, we propose ManiLong-Shot, a novel framework that enables effective OSIL for long-horizon prehensile manipulation tasks. ManiLong-Shot structures long-horizon tasks around physical interaction events, reframing the problem as sequencing interaction-aware primitives instead of directly imitating continuous trajectories. This primitive decomposition can be driven by high-level reasoning from a vision-language model (VLM) or by rule-based heuristics derived from robot state changes. For each primitive, ManiLong-Shot predicts invariant regions critical to the interaction, establishes correspondences between the demonstration and the current observation, and computes the target end-effector pose, enabling effective task execution. Extensive simulation experiments show that ManiLong-Shot, trained on only 10 short-horizon tasks, generalizes to 20 unseen long-horizon tasks across three difficulty levels via one-shot imitation, achieving a 22.8% relative improvement over the SOTA. Additionally, real-robot experiments validate ManiLong-Shot's ability to robustly execute three long-horizon manipulation tasks via OSIL, confirming its practical applicability.

Citations

• AdaptPNP: Integrating Prehensile and Non-Prehensile Skills for Adaptive Robotic Manipulation
• LaViRA: Language-Vision-Robot Actions Translation for Zero-Shot Vision Language Navigation in Continuous Environments
• RoboHiMan: A Hierarchical Evaluation Paradigm for Compositional Generalization in Long-Horizon Manipulation
• DeCo: Task Decomposition and Skill Composition for Zero-Shot Generalization in Long-Horizon 3D Manipulation
• RoboHorizon: An LLM-Assisted Multi-View World Model for Long-Horizon Robotic Manipulation
• FLIP: Flow-Centric Generative Planning as General-Purpose Manipulation World Model
• Instant Policy: In-Context Imitation Learning via Graph Diffusion
• Keypoint Abstraction using Large Models for Object-Relative Imitation Learning
• Autoregressive Action Sequence Learning for Robotic Manipulation
• GravMAD: Grounded Spatial Value Maps Guided Action Diffusion for Generalized 3D Manipulation
• Trust the PRoC3S: Solving Long-Horizon Robotics Problems with LLMs and Constraint Satisfaction
• One-Shot Imitation Learning with Invariance Matching for Robotic Manipulation
• One-Shot Transfer of Long-Horizon Extrinsic Manipulation Through Contact Retargeting
• Point Transformer V3: Simpler, Faster, Stronger
• VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models
• A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondence
• Code as Policies: Language Model Programs for Embodied Control
• Learning Dense Visual Descriptors using Image Augmentations for Robot Manipulation Tasks
• Graph Inverse Reinforcement Learning from Diverse Videos
• TransGrasp: Grasp Pose Estimation of a Category of Objects by Transferring Grasps from Only One Labeled Instance
• Prompting Decision Transformer for Few-Shot Policy Generalization
• Demonstrate Once, Imitate Immediately (DOME): Learning Visual Servoing for One-Shot Imitation Learning
• Lepard: Learning partial point cloud matching in rigid and deformable scenes
• Accelerating Robotic Reinforcement Learning via Parameterized Action Primitives
• Towards More Generalizable One-shot Visual Imitation Learning
• Geometric and Physical Quantities Improve E(3) Equivariant Message Passing
• CaTGrasp: Learning Category-Level Task-Relevant Grasping in Clutter from Simulation
• Coarse-to-Fine Q-attention: Efficient Learning for Visual Robotic Manipulation via Discretisation
• Q-attention: Enabling Efficient Learning for Vision-based Robotic Manipulation
• On the Benefits of Invariance in Neural Networks
• Learning One-Shot Imitation from Humans without Humans
• RLBench: The Robot Learning Benchmark & Learning Environment
• Neural Task Graphs: Generalizing to Unseen Tasks from a Single Video Demonstration
• Dense Object Nets: Learning Dense Visual Object Descriptors By and For Robotic Manipulation
• One-Shot Imitation from Observing Humans via Domain-Adaptive Meta-Learning
• One-Shot Visual Imitation Learning via Meta-Learning
• One-Shot Imitation Learning
• Reinforcement Learning with Parameterized Actions
• Reducing the Barrier to Entry of Complex Robotic Software: a MoveIt! Case Study

Related

• MimicPlay: Long-Horizon Imitation Learning by Watching Human Play
• Towards More Generalizable One-shot Visual Imitation Learning
• One-Shot Robust Imitation Learning for Long-Horizon Visuomotor Tasks from Unsegmented Demonstrations
• Coarse-to-Fine Imitation Learning: Robot Manipulation from a Single Demonstration
• Annotation-Free One-Shot Imitation Learning for Multi-Step Manipulation Tasks