TeleCo: Enabling Whole-Body Teleoperation for High-Precision Data Collection

Data collection is essential for robots to learn complex tasks through human imitation, as it directly impacts model generalization and learning performance. However, existing data collection methods are often constrained to fixed environments and predominantly rely on end-effector pose mapping of robotic arms, resulting in high computational costs, significant latency, and a lack of consideration for whole-body motion strategies. To overcome these limitations, this paper introduces TeleCo, a highly flexible and remotely operable teleoperation data collection system. TeleCo employs a leader-follower architecture and achieves high-precision motion synchronization through a joint matching mechanism. Experimental results show that TeleCo surpasses existing virtual reality (VR)-based approaches in both accuracy and real-time responsiveness while generating high-quality demonstration data to support imitation learning effectively. Additionally, we use data collected by TeleCo to train imitation learning models and validate their effectiveness in robotic manipulation tasks.