TraceVLA Dataset
收藏资源简介:
TraceVLA数据集由马里兰大学学院市分校和微软研究院共同创建,包含15万条机器人操作轨迹,旨在提升视觉语言动作模型在空间-时间动态任务中的表现。该数据集通过视觉轨迹提示技术,将机器人历史动作轨迹编码为视觉输入,帮助模型更好地理解空间和时间关系。数据集的创建过程涉及使用Co-Tracker算法生成密集点轨迹,并将其应用于机器人操作任务中。该数据集主要应用于机器人操作任务的泛化能力提升,旨在解决复杂操作任务中的空间-时间动态问题。
The TraceVLA dataset was co-developed by the University of Maryland, College Park and Microsoft Research, containing 150,000 robotic manipulation trajectories. Its primary goal is to improve the performance of vision-language-action models on spatiotemporal dynamic tasks. By adopting visual trajectory prompting techniques, this dataset encodes historical robotic action trajectories into visual inputs, helping models better understand spatial and temporal relationships. The development of TraceVLA involves employing the Co-Tracker algorithm to generate dense point trajectories, which are then deployed for robotic manipulation tasks. This dataset is mainly applied to enhance the generalization capability of robotic manipulation systems, aiming to resolve spatiotemporal dynamic challenges in complex manipulation scenarios.

- 1TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies马里兰大学学院市分校, 微软研究院 · 2024年



