Details on X-Mobility Training Dataset

[whhuang]

I'd like to generate datasets in the same format as the X-Mobility training datasets. Is the code for generating the X-Mobility datasets using Isaac Sim available anywhere? Otherwise, would it be possible to get a breakdown of all the units/real-world definitions of all the fields in the X-Mobility datasets so they can be replicated? The columns I see are as follows:

['acqtime', 'camera_intrinsic_matrix', 'camera_extrinsic_matrix', 'perspective_semantic_image', 'perspective_semantic_image_shape', 'perspective_semantic_image_labels', 'camera_image', 'driving_command', 'route', 'ego_speed', 'path', 'route_poses', 'goal_id', 'semantic_labels']

In particular, driving_command is a len 5 array, but it appears that only the first and last fields are ever populated. I believe the Nova Carter accepts commands in [linear_speed (m/s), angular_speed (rad/s)], then is the correct mapping for driving_command [linear_speed (m/s), 0, 0, 0, angular_speed (rad/s)]?

Similarly, is ego_speed the euclidean velocity of the vehicle (in m/s)?

The remaining fields appear to map reasonably to MobilityGen, but if there are other fields that have special formats/units, any details would be greatly appreciated. Thank you!

[nv-liuw]

Hi @whhuang,

Great questions! Due to some license issues, we are not able to release the Nav2 with Isaac Sim pipeline, but the MobilityGen can be a great alternative.

For the unit, all fields should follow the international system of units, like m for distance, m/s or rad/s for velocities.

For the driving_command, it should be a len 6 array [vx, vy, vz, ang_x, ang_y, ang_z], and you can adapt to what you robot supports.