DROID
收藏资源简介:
DROID数据集由多个研究机构创建,专注于机器人操作任务的大规模数据收集。该数据集包含36,000条机器人轨迹,涵盖了多种复杂的操作场景和任务。数据集的创建过程结合了机器人自身的感知状态和动作数据,通过引入新的对比损失函数和动作预测损失来增强数据集的质量。DROID数据集主要应用于机器人视觉表示的预训练,旨在提高机器人操作任务的成功率和效率。
The DROID dataset, developed by multiple research institutions, focuses on large-scale data collection for robotic manipulation tasks. This dataset contains 36,000 robotic trajectories, covering a wide range of complex manipulation scenarios and tasks. The development of the dataset integrates the robot's own perceptual states and motion data, and enhances its quality by introducing novel contrastive loss functions and action prediction losses. The DROID dataset is primarily applied to pre-training robotic visual representations, aiming to improve the success rate and efficiency of robotic manipulation tasks.




