RoboTwin
收藏资源简介:
RoboTwin数据集由香港大学和AgileX Robotics合作创建,旨在解决双臂机器人及其工具使用能力的训练数据稀缺问题。该数据集结合了真实世界的遥操作数据和数字孪生生成的合成数据,特别针对双臂机器人工具使用和人与机器人交互的场景。数据集通过COBOT Magic平台收集,包含多种典型任务的数据,并通过AI生成的内容将2D图像转换为详细的3D模型,利用大型语言模型生成专家级训练数据。RoboTwin数据集的应用领域广泛,包括制造业、医疗中心和家庭环境中的机器人操作,旨在提升机器人在复杂任务中的灵活性和效率。
RoboTwin dataset was co-created by The University of Hong Kong and AgileX Robotics, aiming to address the scarcity of training data for dual-arm robots and their tool-use capabilities. This dataset combines real-world teleoperation data and synthetic data generated via digital twins, specifically targeting scenarios involving dual-arm robot tool use and human-robot interaction. The dataset is collected through the COBOT Magic platform, containing data from multiple typical tasks, converts 2D images into detailed 3D models using AI-generated content, and generates expert-level training data with large language models. The RoboTwin dataset has broad application scenarios, including robot manipulation in manufacturing, medical centers and home environments, and aims to enhance the flexibility and efficiency of robots in complex tasks.

- 1RoboTwin: Dual-Arm Robot Benchmark with Generative Digital Twins (early version)香港大学 · 2024年



