design teleoperation interfaces

Novice operators face significant challenges in teleoperating robots, including high operational complexity, poor safety guarantees, and limited cross-platform compatibility—factors that hinder both robot learning and efficient data collection. To address these issues, we propose a low-cost (<$1,000) teleoperation system featuring a novel real-time virtual arm visualization mechanism that enables pre-execution rehearsal of commands, supporting seamless toggling between action preview and execution. The system leverages lightweight motion mapping, real-time virtual rendering, and low-latency interactive feedback, requiring no specialized hardware and ensuring compatibility with mainstream robotic arm platforms. Evaluated on five dexterous manipulation tasks, our approach outperforms existing methods in task success rate and operator efficiency. It substantially reduces the learning curve for novice users and improves operational safety. All source code and deployment documentation are publicly released under an open-source license.

This study addresses the critical challenges of low-quality teleoperation data and poor device-controller compatibility, which hinder the training of embodied intelligence foundation models. To this end, we propose a unified teleoperation data collection framework. Methodologically, we design a quadratic programming (QP)-based optimal controller that integrates dynamic null-space projection and impedance tracking, with adaptive weight tuning to enable joint manipulability-aware compliant control and singularity avoidance. Our end-to-end pipeline unifies position-based inverse kinematics, torque-based inverse dynamics, optimization-based admittance control, and the QP framework. Extensive experiments across diverse robot-device–controller configurations quantitatively evaluate trajectory tracking error, singularity occurrence rate, and joint motion smoothness. Results demonstrate substantial improvements in data stability and diversity, yielding a high-quality, broad-coverage robotic skill dataset essential for advancing embodied intelligence.

Existing teleoperation methods suffer from strong hardware dependencies and mismatched human–robot control frequencies, limiting their applicability in hazardous environment operations and robot learning demonstration collection. To address these challenges, this paper proposes a cross-platform plug-and-play teleoperation framework. Our approach introduces: (i) a universal hardware interface automatically derived from URDF models; (ii) an online continuous trajectory generation algorithm that operates without low-level closed-loop access; (iii) minimum-stretch spline optimization for enhanced motion smoothness; and (iv) dynamic switching between precision-priority and velocity-priority control modes. Implemented with a C++ core and Python API, the system demonstrates broad generalizability, real-time performance, and smooth control across multiple single- and dual-arm robotic platforms. It achieves seamless mapping from low-frequency human inputs to high-frequency robot execution, significantly improving operational flexibility and deployment efficiency.

In robotic imitation learning, acquiring high-quality demonstration data remains costly, inefficient, and lacking in force feedback support. To address this, we propose an open-source, low-cost, force-feedback-enabled joint-space mapping teleoperation system specifically designed for UR manipulators and supporting both single- and dual-arm collaborative task demonstrations. Our key contributions are: (1) a novel lightweight adaptive-sensitivity force-feedback controller that jointly optimizes operational intuitiveness, trajectory accuracy, and demonstration reproducibility; (2) a modular architecture based on joint-space mapping and ROS integration, enabling rapid adaptation to diverse manipulators and humanoid robots; and (3) a custom force-controlled hand controller with a real-time synchronized data recording interface. Experiments demonstrate significant improvements in data collection efficiency and annotation quality. All hardware designs, assembly instructions, and software code are publicly released under an open-source license and have been widely adopted by the research community.

Existing learning-from-demonstration approaches are largely constrained to a single teaching modality—such as teleoperation, kinesthetic teaching, or natural demonstration—limiting their ability to accommodate diverse human demonstrators’ preferences and task requirements. This paper introduces the Versatile Demonstration Interface (VDI), a unified, hardware-efficient teaching interface designed for industrial collaborative robotics. VDI is the first solution to natively support all three modalities on a single platform without requiring environment calibration. Its multimodal perception architecture fuses AprilTag-based visual tracking, six-axis force sensing, and joint encoder/external pose estimation, balancing robustness with human-centered ergonomics. An expert user study conducted at a local manufacturing innovation center validates VDI’s effectiveness across representative industrial tasks. Furthermore, the study yields multiple production-line application patterns, demonstrating practical scalability and adaptability in real-world settings.

Existing teleoperation systems are often tailored to specific hardware and tasks, lacking generality and scalability. This work proposes ModPack—a modular and extensible teleoperation framework that leverages a wearable backpack platform integrating computation, power, communication, and storage functionalities, coupled with a unified wearable interface architecture. This design enables plug-and-play integration of capability modules such as joint-level teleoperation, mobile manipulation, and active perception. The system’s efficacy in real-world mobile manipulation tasks is validated across two heterogeneous robotic platforms, demonstrating significantly enhanced reusability and adaptability. To foster community advancement, the authors open-source all hardware and software designs.

为解决高自由度手部操作数据收集难题，DITTO通过设计7-DOF机械手和动力外骨骼，实现了带有力反馈的双侧遥操作与自然数据采集。

This study addresses the limitations of conventional joystick-based teleoperation in scenarios such as nuclear industry applications, where precise path tracking, force control, and obstacle avoidance during surface contact tasks are critical, yet impose high cognitive load on operators. To overcome these challenges, this work proposes a touchscreen-based teleoperation interface that directly maps continuous finger motion to the Franka Emika Panda robotic arm, integrating control and visualization for more intuitive motion mapping and fine-grained velocity modulation. User studies demonstrate that, compared to both joystick and one-button autonomous modes, the proposed method reduces median task completion time by 53.5% (2.50 vs. 5.38 minutes), achieves a 90.7% coverage rate on sinusoidal paths with lower overshoot, and decreases NASA-TLX cognitive workload scores by 17.3%, significantly enhancing operational efficiency and naturalness.

This work addresses the limitations of handheld data collection—which captures only human observation actions and lacks dynamic feasibility during contact-sensitive phases—and the high cost of full-task teleoperation. To bridge this gap, the authors propose BRIDGE, a method that combines low-cost handheld data with a small number of staged teleoperated demonstrations. BRIDGE employs a state-gated mixture-of-experts architecture that dynamically routes the robot’s current state to the appropriate task-phase-specific expert policy, enabling phase-adaptive fusion of observed and desired actions. By integrating diffusion-based policy experts with a state-conditioned gating mechanism, BRIDGE achieves significant performance gains: across three contact-intensive manipulation tasks, it improves task success rates by up to 36.7% compared to a baseline using only handheld data.

Fixed Cartesian impedance control struggles to balance task performance and safety in contact-intensive teleoperation, as its gains concurrently govern task progression, contact support, and force/impact characteristics. This work proposes a Controller-to-Manifold impedance redirection method (C2M) based on a single demonstration, which decomposes the task into a manifold component and a passive residual model to generate variable impedance commands while preserving the semantic content of the demonstrated trajectory. Integrated with Manifold-constrained Parameter Optimization (MPO), the approach substantially reduces force peaks, impulse, force variability, and controller energy consumption. Fifteen closed-loop experiments on a Franka Panda platform demonstrate successful completion of all tasks, with all four aggressiveness metrics significantly outperforming those of the original fixed-impedance strategy.