RoboMatch: A Unified Mobile-Manipulation Teleoperation Platform with Auto-Matching Network Architecture for Long-Horizon Tasks

Abstract

This paper presents RoboMatch, a novel unified teleoperation platform for mobile manipulation with an auto-matching network architecture, designed to tackle long-horizon tasks in dynamic environments. Our system enhances teleoperation performance, data collection efficiency, task accuracy, and operational stability. The core of RoboMatch is a cockpit-style control interface that enables synchronous operation of the mobile base and dual arms, significantly improving control precision and data collection. Moreover, we introduce the Proprioceptive-Visual Enhanced Diffusion Policy (PVE-DP), which leverages Discrete Wavelet Transform (DWT) for multi-scale visual feature extraction and integrates high-precision IMUs at the end-effector to enrich proprioceptive feedback, substantially boosting fine manipulation performance. Furthermore, we propose an Auto-Matching Network (AMN) architecture that decomposes long-horizon tasks into logical sequences and dynamically assigns lightweight pre-trained models for distributed inference. Experimental results demonstrate that our approach improves data collection efficiency by over 20%, increases task success rates by 20-30% with PVE-DP, and enhances long-horizon inference performance by approximately 40% with AMN, offering a robust solution for complex manipulation tasks. Project website: https://robomatch.github.io

Figures (6)

• Figure 1: Overview of the RoboMatch framework. This figure integrates three core components: (1) RoboMatch, a unified mobile-manipulation teleoperation platform that matches the robot base with the operation platform to achieve high-precision master-slave control and immersive observation; (2) PVE-DP, a policy combining spatio-frequency visual enhancement and rich proprioception to improve fine manipulation accuracy; (3) AMN, an architecture that combines the semantic parsing capabilities of VLMs with the execution advantages of specialized small policy networks for long-horizon execution, enabling chain-of-thought reasoning for complex task decomposition and adaptive operation.
• Figure 2: RoboMatch teleoperation demonstration.
• Figure 3: Overall network architecture of PVE-DP, which consists of three core modules: (1) FE-EMA Module enhances robotic visual perception by integrating spatio-frequency features from visual data; (2) Rich Proprioception strengthens the robot's self-state awareness through concatenating arm joint positions (Qpos) and end-effector quaternion (Quat) from IMU sensing; (3) DP U-Net takes the enhanced visual-proprioceptive observations as conditions and predicts fine-grained action noise to achieve precise manipulation during robot inference.
• Figure 4: Overview of AMN framework. This figure presents the Auto-Matching Network architecture, featuring: (1) Task Decomposition of complex task $\mathcal{T}$ into sub-tasks for specialized policy networks; (2) Vision-Language Input processing both linguistic commands and visual scenes; (3) Chain-of-Thought Thinking and Planning that breaks tasks into sequential steps matched with pre-trained policies; (4) Inference Execution using PVE-DP with wrist quaternion and visual spatio-frequency fusion for precise manipulation.
• Figure 5: Comparison between DP and the proposed PVE-DP method in simulation and real-world tasks. The first row demonstrates the inference results of DP, while the second row presents the improved performance of PVE-DP.