Video Prediction for Visual Planning (VP2)
收藏资源简介:
VP2数据集由斯坦福大学创建,旨在为视频预测模型提供一个控制基准,以评估模拟机器人操作中的性能。该数据集包括模拟环境,具体任务实例规范,以及包含每个任务类别脚本交互轨迹的训练数据集。VP2数据集的设计目标是提供一个简单的接口,使得几乎任何动作条件下的视频预测模型都可以直接评估。数据集的应用领域主要集中在机器人操作的规划问题上,旨在通过视频预测模型提高机器人在多任务环境中的操作成功率。
The VP2 dataset was created by Stanford University, with the aim of providing a controlled benchmark for video prediction models to evaluate their performance in simulated robotic manipulation scenarios. This dataset includes simulated environments, specifications for specific task instances, and training datasets that contain scripted interactive trajectories for each task category. The VP2 dataset is designed to offer a simple interface, enabling direct evaluation of almost any action-conditioned video prediction model. The primary application areas of this dataset focus on planning problems in robotic manipulation, aiming to improve the operational success rates of robots in multi-task environments via video prediction models.




