VinT-6D
收藏资源简介:
VinT-6D是由腾讯机器人X实验室、中山大学等机构联合创建的大规模多模态数据集,旨在提升机器人手内物体姿态估计的精度。该数据集包含200万条模拟数据(VinT-Sim)和10万条真实数据(VinT-Real),涵盖了视觉、触觉和本体感知信息。数据集通过MuJoCo和Blender进行模拟生成,并通过定制的多模态机器人平台收集真实数据,确保了数据的高质量和多模态对齐。VinT-6D主要用于机器人手内操作任务,特别是在视觉被遮挡的情况下,通过融合触觉和本体感知信息来提升物体姿态估计的准确性。
VinT-6D is a large-scale multimodal dataset jointly created by Tencent Robotics X Lab, Sun Yat-sen University and other institutions, aiming to improve the accuracy of robotic in-hand object pose estimation. It contains 2 million simulated data entries (VinT-Sim) and 100,000 real-world data entries (VinT-Real), covering visual, tactile and proprioceptive information. The dataset is simulated and generated via MuJoCo and Blender, while real-world data is collected using a custom multimodal robotic platform, ensuring high data quality and multimodal alignment. VinT-6D is mainly used for robotic in-hand manipulation tasks, especially to improve the accuracy of object pose estimation by fusing tactile and proprioceptive information under visual occlusion conditions.




