Terrain-Robustness Benchmark
收藏资源简介:
Terrain-Robustness Benchmark是由苏黎世联邦理工学院和加州理工学院合作创建的数据集,旨在为足式机器人的地形感知运动提供一个多样化和具有挑战性的地形数据集。该数据集包含256个地形样本,这些样本是通过地形创作和主动学习方法生成的,以模拟真实世界中的非结构化地形。数据集的创建过程涉及使用条件生成对抗网络(GANs)和马尔可夫决策过程(MDP)来生成高质量的地形样本。该数据集的应用领域主要集中在评估和提高足式机器人在复杂地形上的运动鲁棒性,解决机器人在自然环境中运动时的挑战。
Terrain-Robustness Benchmark is a dataset collaboratively developed by ETH Zurich and the California Institute of Technology, designed to provide a diverse and challenging terrain dataset for terrain-aware locomotion of legged robots. This dataset comprises 256 terrain samples generated via terrain generation and active learning methods to simulate unstructured terrain in real-world scenarios. The dataset creation process utilizes conditional Generative Adversarial Networks (GANs) and Markov Decision Processes (MDPs) to produce high-quality terrain samples. Its primary applications focus on evaluating and enhancing the locomotion robustness of legged robots on complex terrain, and addressing the challenges encountered by robots during movement in natural environments.




