kmax2001/real_task3_rlds

RLDS (TFDS TFRecord) export of the LeRobot dataset real_task3 recorded on an SO101 follower arm with two Intel RealSense D435i cameras (front, right, 640x360 @ 30 fps).

Converted with lerobot2rlds_2cam_factoryK.py (kmax2001/lerobot2rlds).

Layout

real_task3/0.1.0/
  dataset_info.json
  features.json
  real_task3-train.tfrecord-0000X-of-0000N

Step schema

key	shape	note
observation/front, observation/right	(360, 640, 3) uint8	RGB, JPEG encoded
observation/joint_state	(6,) float32	shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper
observation/state	(7,) float32	end-effector x, y, z, wx, wy, wz, gripper_pos
observation/cam_front_intrinsic, observation/cam_right_intrinsic	(3, 3) float32	RealSense factory K at 640x360
action	(6,) float32	target joint positions
language_instruction	text
reward / discount	float32	0.0 / 1.0
is_first / is_last / is_terminal	bool

Camera extrinsics are not included. The factory distortion coefficients of both cameras are all zero, so no undistortion was applied.

Load

import tensorflow_datasets as tfds
b = tfds.builder_from_directory("<local path>/real_task3/0.1.0")
ds = b.as_dataset(split="train")