n5zhong/libero-100-random-traces

这是一个修改版的LIBERO-100数据集，基于yilin-wu的数据集进行扩展。数据集包含修改后的技能注释，这些注释提供了轨迹数据，用于训练轨迹生成器。通过LIBERO模拟器，额外添加了4,000条轨迹，这些轨迹将机器人放置在某个起始位置，并沿着记录的演示轨迹移动，本质上是一些无意义的移动命令。其目的是训练机器人控制器遵循轨迹，避免对特定场景过拟合，从而提升泛化能力。A modified version of the LIBERO-100 dataset, forked from yilin-wus dataset. The modified skill annotations contain trace data for training a trace generator. Using the LIBERO simulator, an additional 4,000 trajectories have been added, which start the robot at a given position and move it along a trace determined by a recorded demo in the dataset. These are essentially nonsense move commands, intended to train the controller to obey traces and prevent overfitting to specific scenes.