Datasets:
Head Camera Motion and Full Task Recording
Hi, I have a couple of questions about the data collection setup.
First, during the grasping task, was the eye camera kept fixed, or did the robot’s head move? If the head was moving, were the head or neck joint states also recorded and included in the training data? Since the eye cameras move together with the head, I assume the model may need the head motion as part of the observation or action space.
Second, I noticed that the dataset mainly contains the tabletop grasping segment. Was only the period from approaching the object to completing the grasp recorded? For a complete manipulation episode, would it not be better to include the full sequence, such as starting from the initial raised-arm pose, reaching towards the object, grasping it, lifting it, and then placing it down or returning to the initial pose?
I would appreciate any clarification about how the episode boundaries were defined and why only the tabletop grasping segment was included.