PsiBotAI/SynData
收藏资源简介:
SynData是由PsiBot新发布的下一代大规模真实世界多模态数据集。它全面覆盖了视觉、语言和动作等关键维度,并提供高度真实、高密度、高可用性的人类数据,为具身智能训练奠定了坚实基础。该数据集基于PsiBot自主研发的专为数据收集设计的外骨骼手套系统,实现了高达毫米级的定位精度,并捕捉了双手和手臂的全部自由度,同时保留了真实世界任务中操作者的自然行为。值得注意的是,SynData既包含基于外骨骼的数据,也包含裸手数据,结合了高精度结构化捕捉与自然人类交互行为,为具身模型提供了更丰富、更全面的学习样本。基于这种紧密集成的硬件-软件数据收集流程,SynData显著提高了数据的一致性、可重用性和泛化价值,并进一步加速了具身智能从实验室研究到真实工业部署的过渡。本版本专为视觉-动作建模、操作学习、模仿学习和多模态具身智能研究而设计。
SynData is a next-generation large-scale real-world multimodal dataset newly released by PsiBot. It comprehensively covers key dimensions including vision, language, and action, and provides highly realistic, high-density, and highly usable human data as a solid foundation for embodied intelligence training. Powered by PsiBots self-developed exoskeleton glove system designed specifically for data collection, SynData achieves up to millimeter-level positioning accuracy and captures the full degrees of freedom of both hands and arms, while preserving natural operator behavior during real-world tasks. Notably, SynData includes both exoskeleton-based data and bare-hand data, combining high-precision structured capture with natural human interaction behavior to provide richer and more comprehensive learning samples for embodied models. Built on this tightly integrated hardware-software data collection pipeline, SynData significantly improves data consistency, reusability, and generalization value, and further accelerates the transition of embodied intelligence from laboratory research to real industrial deployment. This release is designed for vision-action modeling, manipulation learning, imitation learning, and multimodal embodied intelligence research.





