BrandonAL/eval_smolvla_spline_gpu_c100_libero_object

这是一个机器人控制与视觉数据集，专门用于强化学习或模仿学习任务。数据集中包含多个任务（共10个），每个任务代表不同的机器人操作场景。每个数据样本记录了机器人在特定时间步的状态信息：包括两个视角的图像观察（image和image2）、机器人状态向量（34维浮点数）、执行的动作（7维浮点数）及其原始值、即时奖励、任务完成标志（done）和成功标志（success）。此外，还包含元数据如回合索引、帧索引和时间戳。数据集总大小约5.07GB，涵盖超过2.8万个样本，适用于训练和评估机器人控制策略。

This is a robot control and vision dataset designed for reinforcement learning or imitation learning tasks. The dataset includes multiple tasks (10 in total), each representing different robotic manipulation scenarios. Each data sample captures the robots state at a specific timestep: including two-view image observations (image and image2), robot state vector (34-dimensional floats), executed actions (7-dimensional floats) and their raw values, immediate reward, task completion flag (done), and success flag (success). Additionally, metadata such as episode index, frame index, and timestamp are provided. The total dataset size is approximately 5.07GB, covering over 28,000 samples, suitable for training and evaluating robot control policies.