FLAME
收藏资源简介:
FLAME是一个面向机器人操作任务联邦学习的基准,由KTH皇家理工学院创建。该数据集包含超过160万个数据样本,涵盖20种不同的机器人操作任务,每个任务在不同的环境设置下收集了多样化的操作演示。FLAME的特色在于将分布式数据集整合到联邦学习框架中,为评估联邦学习算法提供了一个基础。数据集的创建目的是为了解决机器人操作学习中的可扩展性、适应性和隐私保护问题。
FLAME is a benchmark for federated learning in robotic manipulation tasks, developed by KTH Royal Institute of Technology. This dataset contains over 1.6 million data samples, covering 20 distinct robotic manipulation tasks, with diverse manipulation demonstrations collected for each task across varying environmental settings. A core feature of FLAME is its integration of distributed datasets into the federated learning framework, providing a foundational platform for evaluating federated learning algorithms. The dataset was created to address the key challenges of scalability, adaptability, and privacy preservation in robotic manipulation learning.




