AMO数据集
收藏资源简介:
AMO数据集是一个专门为灵活操作任务设计的人形机器人运动数据集,旨在解决人形机器人在现实世界中实现高度灵活的全身运动控制问题。该数据集包含由动作捕捉数据生成的臂部轨迹和随机采样的躯干方向,通过动力学感知轨迹优化器生成全身参考运动,满足运动学和动力学约束。数据集的创建过程包括使用模型基础轨迹优化进行数据收集,以及使用多目标逆运动学进行控制信号的计算。该数据集的应用领域主要在于人形机器人的灵活操作任务,例如从地面拾取物体等。
The AMO dataset is a humanoid robot motion dataset specifically designed for dexterous manipulation tasks, aiming to address the critical challenge of achieving highly flexible full-body motion control for humanoid robots in real-world scenarios. This dataset includes arm trajectories generated from motion capture data and randomly sampled torso orientations, and generates full-body reference motions via dynamics-aware trajectory optimizers that satisfy both kinematic and dynamic constraints. The dataset creation process involves data collection using model-based trajectory optimization, as well as the computation of control signals through multi-objective inverse kinematics. The primary application areas of this dataset focus on dexterous manipulation tasks for humanoid robots, such as picking up objects from the ground.




