现实世界四足运动基准数据集

现实世界四足运动基准数据集是由西湖大学工程学院机器智能实验室创建，旨在为离线强化学习算法提供一个真实的测试平台。该数据集包含30个任务，涉及五种不同地形和六种运动命令，数据通过经典的模型预测控制（MPC）方法收集，而非传统的无模型在线RL方法。数据集的创建过程注重任务的多样性和难度，以真实反映离线RL算法在实际应用中的挑战。该数据集主要应用于四足机器人的稳定控制和快速适应性研究，旨在解决复杂地形下的机器人运动控制问题。

The Real-World Quadruped Locomotion Benchmark Dataset was created by the Machine Intelligence Laboratory, School of Engineering, Westlake University, aiming to provide a realistic testbed for offline reinforcement learning algorithms. This dataset contains 30 tasks covering five distinct terrains and six motion commands, with data collected via the classic Model Predictive Control (MPC) method rather than traditional model-free online RL methods. The dataset development process prioritizes task diversity and difficulty to authentically reflect the challenges faced by offline RL algorithms in real-world applications. This dataset is primarily applied to research on stable control and rapid adaptability of quadruped robots, aiming to address robotic motion control issues in complex terrains.