Long-Horizon Manipulation (LHManip) dataset
收藏资源简介:
Long-Horizon Manipulation (LHManip) 数据集由意大利技术研究院和昆士兰科技大学机器人中心合作创建,包含200个真实机器人远程操作的演示,涉及20种不同的操作任务。数据集中的任务包括抓取、推动、堆叠和投掷物体等,每个任务都配有自然语言指令和多摄像头视角。LHManip数据集旨在解决长视野、多步骤的机器人操作任务,支持算法在多样场景中的泛化能力,适应新环境,并解决需要多步骤执行的任务挑战。数据集的应用领域包括家庭机器人的日常任务,如餐食准备、房间清洁和空间组织,强调了机器人系统中感知技能的重要性。
The Long-Horizon Manipulation (LHManip) dataset was co-developed by the Italian Institute of Technology and the Robotics Centre of Queensland University of Technology. It contains 200 real robotic teleoperation demonstrations covering 20 distinct manipulation tasks, including object grasping, pushing, stacking, and throwing. Each task is paired with natural language instructions and multi-camera viewpoints. The LHManip dataset aims to address long-horizon, multi-step robotic manipulation tasks, supporting the generalization capability of algorithms across diverse scenarios, enabling adaptation to novel environments, and tackling the challenges associated with multi-step task execution. Its application domains include daily tasks for home robots such as meal preparation, room cleaning, and spatial organization, highlighting the importance of perceptual skills in robotic systems.

- 1LHManip: A Dataset for Long-Horizon Language-Grounded Manipulation Tasks in Cluttered Tabletop Environments意大利技术研究院 (IIT), 热那亚, 意大利 · 2023年



