LIBERO-Unseen
收藏资源简介:
LIBERO-Unseen数据集是LIBERO-90数据集的修改版本,用于评估机器人在未见任务上的泛化能力。该数据集包含40个未见任务,旨在通过模拟环境中的机器人操作来测试和提升机器人在面对新任务时的适应性和成功率。数据集的创建是为了解决机器人视觉-语言-动作模型在处理未见任务时存在的泛化问题,通过提供一系列未见任务,研究人员可以评估其模型在不同场景下的表现。
The LIBERO-Unseen dataset is a modified variant of the LIBERO-90 dataset, developed to evaluate a robot's generalization capability on unseen tasks. It comprises 40 unseen tasks, with the goal of testing and enhancing a robot's adaptability and success rate when encountering novel tasks via robotic manipulations in simulated environments. This dataset was created to address the generalization issues faced by robotic vision-language-action models when handling unseen tasks. By providing a series of such unseen tasks, researchers can assess their models' performance across diverse scenarios.

- 13D CAVLA: Leveraging Depth and 3D Context to Generalize Vision Language Action Models for Unseen Tasks纽约大学 · 2025年



