Open-X-Embodiment

该数据集名为Open-X-Embodiment，包含了97万条真实世界的机器人演示数据。该数据集被用于对OpenVLA模型进行预训练，并作为评估包括NORA在内的各种视觉语言动作模型的基础，其任务旨在训练和评估针对机器人任务所需的视觉、语言和动作模型。The dataset named Open-X-Embodiment consists of 970,000 real-world robotic demonstration samples. It is utilized for pre-training the OpenVLA model and serves as the foundational benchmark for evaluating various vision-language-action models including NORA, with the core objective of supporting the training and evaluation of vision, language, and action models required for robotic tasks.

搜集汇总

背景与挑战

背景概述

Open-X-Embodiment是一个大规模机器人数据集，包含97万条真实世界演示数据，主要用于OpenVLA模型的预训练和视觉语言动作模型的评估。该数据集的核心目标是支持机器人任务中视觉、语言和动作模型的训练与测试。