PlasticineLab

PlasticineLab是由加州大学圣地亚哥分校等机构开发的一个包含50个配置的软体操作基准数据集。该数据集专注于模拟弹性与塑性材料的变形，通过不同的任务如捏合、滚动、切割等，来评估机器学习算法在软体操作上的性能。数据集利用可微分物理引擎，首次在软体基准中提供分析梯度信息，支持基于梯度的优化和监督学习。PlasticineLab的应用领域包括机器人学、计算机图形学和材料科学，旨在解决复杂物理环境下软体操作的技能学习问题。PlasticineLab is a benchmark dataset for soft-body manipulation containing 50 configurations, developed by institutions including the University of California, San Diego and others. This dataset focuses on simulating the deformation of elastic and plastic materials, and evaluates the performance of machine learning algorithms for soft-body manipulation via various tasks such as pinching, rolling, cutting and so on. Leveraging a differentiable physics engine, it is the first soft-body manipulation benchmark to provide analytical gradient information, supporting gradient-based optimization and supervised learning. The application domains of PlasticineLab cover robotics, computer graphics and materials science, aiming to address the skill learning challenges for soft-body manipulation in complex physical environments.