Automated documentation update.

PiperOrigin-RevId: 670284563

Author: The TensorFlow Datasets Authors

docs/catalog/_toc.yaml

@@ -610,6 +610,9 @@ toc:
     title: databricks_dolly
   - path: /datasets/catalog/dices
     title: dices
+  - path: /datasets/catalog/dolma
+    status: nightly
+    title: dolma
   - path: /datasets/catalog/e2e_cleaned
     title: e2e_cleaned
   - path: /datasets/catalog/irc_disentanglement
@@ -668,6 +671,9 @@ toc:
     title: bot_adversarial_dialogue
   - path: /datasets/catalog/dices
     title: dices
+  - path: /datasets/catalog/dolma
+    status: nightly
+    title: dolma
   - path: /datasets/catalog/e2e_cleaned
     title: e2e_cleaned
   - path: /datasets/catalog/imdb_reviews
@@ -1358,6 +1364,9 @@ toc:
     title: dices
   - path: /datasets/catalog/doc_nli
     title: doc_nli
+  - path: /datasets/catalog/dolma
+    status: nightly
+    title: dolma
   - path: /datasets/catalog/dolphin_number_word
     title: dolphin_number_word
   - path: /datasets/catalog/drop

docs/catalog/asu_table_top_converted_externally_to_rlds.md

@@ -48,31 +48,31 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(7,), dtype=float32),
-        'action_delta': Tensor(shape=(7,), dtype=float32),
+        'action': Tensor(shape=(7,), dtype=float32, description=Robot action, consists of [7x joint velocities, 2x gripper velocities, 1x terminate episode].),
+        'action_delta': Tensor(shape=(7,), dtype=float32, description=Robot delta action, consists of [7x joint velocities, 2x gripper velocities, 1x terminate episode].),
         'action_inst': Text(shape=(), dtype=string),
-        'discount': Scalar(shape=(), dtype=float32),
+        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
         'goal_object': Text(shape=(), dtype=string),
         'ground_truth_states': FeaturesDict({
-            'EE': Tensor(shape=(6,), dtype=float32),
-            'bottle': Tensor(shape=(6,), dtype=float32),
-            'bread': Tensor(shape=(6,), dtype=float32),
-            'coke': Tensor(shape=(6,), dtype=float32),
-            'cube': Tensor(shape=(6,), dtype=float32),
-            'milk': Tensor(shape=(6,), dtype=float32),
-            'pepsi': Tensor(shape=(6,), dtype=float32),
+            'EE': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'bottle': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'bread': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'coke': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'cube': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'milk': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
+            'pepsi': Tensor(shape=(6,), dtype=float32, description=xyzrpy),
         }),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'image': Image(shape=(224, 224, 3), dtype=uint8),
-            'state': Tensor(shape=(7,), dtype=float32),
-            'state_vel': Tensor(shape=(7,), dtype=float32),
+            'image': Image(shape=(224, 224, 3), dtype=uint8, description=Main camera RGB observation.),
+            'state': Tensor(shape=(7,), dtype=float32, description=Robot state, consists of [6x robot joint angles, 1x gripper position].),
+            'state_vel': Tensor(shape=(7,), dtype=float32, description=Robot joint velocity, consists of [6x robot joint angles, 1x gripper position].),
         }),
-        'reward': Scalar(shape=(), dtype=float32),
+        'reward': Scalar(shape=(), dtype=float32, description=Reward if provided, 1 on final step for demos.),
     }),
 })
 ```

docs/catalog/austin_buds_dataset_converted_externally_to_rlds.md

@@ -48,19 +48,19 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(7,), dtype=float32),
-        'discount': Scalar(shape=(), dtype=float32),
+        'action': Tensor(shape=(7,), dtype=float32, description=Robot action, consists of [6x end effector delta pose, 1x gripper position].),
+        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'image': Image(shape=(128, 128, 3), dtype=uint8),
-            'state': Tensor(shape=(24,), dtype=float32),
-            'wrist_image': Image(shape=(128, 128, 3), dtype=uint8),
+            'image': Image(shape=(128, 128, 3), dtype=uint8, description=Main camera RGB observation.),
+            'state': Tensor(shape=(24,), dtype=float32, description=Robot state, consists of [7x robot joint angles, 1x gripper position, 16x robot end-effector homogeneous matrix].),
+            'wrist_image': Image(shape=(128, 128, 3), dtype=uint8, description=Wrist camera RGB observation.),
         }),
-        'reward': Scalar(shape=(), dtype=float32),
+        'reward': Scalar(shape=(), dtype=float32, description=Reward if provided, 1 on final step for demos.),
     }),
 })
 ```

docs/catalog/austin_sailor_dataset_converted_externally_to_rlds.md

@@ -48,22 +48,22 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(7,), dtype=float32),
-        'discount': Scalar(shape=(), dtype=float32),
+        'action': Tensor(shape=(7,), dtype=float32, description=Robot action, consists of [3x ee relative pos, 3x ee relative rotation, 1x gripper action].),
+        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'image': Image(shape=(128, 128, 3), dtype=uint8),
-            'state': Tensor(shape=(8,), dtype=float32),
-            'state_ee': Tensor(shape=(16,), dtype=float32),
-            'state_gripper': Tensor(shape=(1,), dtype=float32),
-            'state_joint': Tensor(shape=(7,), dtype=float32),
-            'wrist_image': Image(shape=(128, 128, 3), dtype=uint8),
+            'image': Image(shape=(128, 128, 3), dtype=uint8, description=Main camera RGB observation.),
+            'state': Tensor(shape=(8,), dtype=float32, description=Default robot state, consists of [3x robot ee pos, 3x ee quat, 1x gripper state].),
+            'state_ee': Tensor(shape=(16,), dtype=float32, description=End-effector state, represented as 4x4 homogeneous transformation matrix of ee pose.),
+            'state_gripper': Tensor(shape=(1,), dtype=float32, description=Robot gripper opening width. Ranges between ~0 (closed) to ~0.077 (open)),
+            'state_joint': Tensor(shape=(7,), dtype=float32, description=Robot 7-dof joint information (not used in original SAILOR dataset).),
+            'wrist_image': Image(shape=(128, 128, 3), dtype=uint8, description=Wrist camera RGB observation.),
         }),
-        'reward': Scalar(shape=(), dtype=float32),
+        'reward': Scalar(shape=(), dtype=float32, description=True on last step of the episode.),
     }),
 })
 ```

docs/catalog/austin_sirius_dataset_converted_externally_to_rlds.md

@@ -48,24 +48,24 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(7,), dtype=float32),
-        'action_mode': Tensor(shape=(1,), dtype=float32),
-        'discount': Scalar(shape=(), dtype=float32),
-        'intv_label': Tensor(shape=(1,), dtype=float32),
+        'action': Tensor(shape=(7,), dtype=float32, description=Robot action, consists of [3x ee relative pos, 3x ee relative rotation, 1x gripper action].),
+        'action_mode': Tensor(shape=(1,), dtype=float32, description=Type of interaction. -1: initial human demonstration. 1: intervention. 0: autonomuos robot execution (includes pre-intervention class)),
+        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
+        'intv_label': Tensor(shape=(1,), dtype=float32, description=Same as action_modes, except 15 timesteps preceding intervention are labeled as -10.),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'image': Image(shape=(84, 84, 3), dtype=uint8),
-            'state': Tensor(shape=(8,), dtype=float32),
-            'state_ee': Tensor(shape=(16,), dtype=float32),
-            'state_gripper': Tensor(shape=(1,), dtype=float32),
-            'state_joint': Tensor(shape=(7,), dtype=float32),
-            'wrist_image': Image(shape=(84, 84, 3), dtype=uint8),
+            'image': Image(shape=(84, 84, 3), dtype=uint8, description=Main camera RGB observation.),
+            'state': Tensor(shape=(8,), dtype=float32, description=Default robot state, consists of [7x robot joint state, 1x gripper state].),
+            'state_ee': Tensor(shape=(16,), dtype=float32, description=End-effector state, represented as 4x4 homogeneous transformation matrix of ee pose.),
+            'state_gripper': Tensor(shape=(1,), dtype=float32, description=Robot gripper opening width. Ranges between ~0 (closed) to ~0.077 (open)),
+            'state_joint': Tensor(shape=(7,), dtype=float32, description=Robot 7-dof joint information.),
+            'wrist_image': Image(shape=(84, 84, 3), dtype=uint8, description=Wrist camera RGB observation.),
         }),
-        'reward': Scalar(shape=(), dtype=float32),
+        'reward': Scalar(shape=(), dtype=float32, description=Reward if provided, 1 on final step for demos.),
     }),
 })
 ```

docs/catalog/bc_z.md

@@ -47,23 +47,23 @@ Split     | Examples
 FeaturesDict({
     'steps': Dataset({
         'action': FeaturesDict({
-            'future/axis_angle_residual': Tensor(shape=(30,), dtype=float32),
-            'future/target_close': Tensor(shape=(10,), dtype=int64),
-            'future/xyz_residual': Tensor(shape=(30,), dtype=float32),
+            'future/axis_angle_residual': Tensor(shape=(30,), dtype=float32, description=The next 10 actions for the rotation. Each action is a 3D delta to add to the current axis angle.),
+            'future/target_close': Tensor(shape=(10,), dtype=int64, description=The next 10 actions for the gripper. Each action is the value the gripper closure should be changed to (notably it is *not* a delta.)),
+            'future/xyz_residual': Tensor(shape=(30,), dtype=float32, description=The next 10 actions for the positions. Each action is a 3D delta to add to current position.),
         }),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
         'observation': FeaturesDict({
             'episode_success': float32,
-            'image': Image(shape=(171, 213, 3), dtype=uint8),
-            'natural_language_embedding': Tensor(shape=(512,), dtype=float32),
+            'image': Image(shape=(171, 213, 3), dtype=uint8, description=Camera image of the robot, downsampled 3x),
+            'natural_language_embedding': Tensor(shape=(512,), dtype=float32, description=An embedding of the task via Universal Sentence Encoder (https://tfhub.dev/google/universal-sentence-encoder/4)),
             'natural_language_instruction': string,
             'present/autonomous': int64,
-            'present/axis_angle': Tensor(shape=(3,), dtype=float32),
+            'present/axis_angle': Tensor(shape=(3,), dtype=float32, description=The current rotation of the end effector in axis-angle representation.),
             'present/intervention': int64,
-            'present/sensed_close': Tensor(shape=(1,), dtype=float32),
-            'present/xyz': Tensor(shape=(3,), dtype=float32),
+            'present/sensed_close': Tensor(shape=(1,), dtype=float32, description=How much the gripper is currently closed. Scaled from 0 to 1, but not all values from 0 to 1 are reachable. The range in the data is about 0.2 to 1),
+            'present/xyz': Tensor(shape=(3,), dtype=float32, description=The current position of the end effector in axis-angle representation, in robot frame),
             'sequence_length': int64,
         }),
         'reward': Scalar(shape=(), dtype=float32),

docs/catalog/berkeley_autolab_ur5.md

@@ -48,9 +48,9 @@ FeaturesDict({
     'steps': Dataset({
         'action': FeaturesDict({
             'gripper_closedness_action': float32,
-            'rotation_delta': Tensor(shape=(3,), dtype=float32),
+            'rotation_delta': Tensor(shape=(3,), dtype=float32, description=Delta change in roll, pitch, yaw.),
             'terminate_episode': float32,
-            'world_vector': Tensor(shape=(3,), dtype=float32),
+            'world_vector': Tensor(shape=(3,), dtype=float32, description=Delta change in XYZ.),
         }),
         'is_first': bool,
         'is_last': bool,
@@ -61,7 +61,7 @@ FeaturesDict({
             'image_with_depth': Image(shape=(480, 640, 1), dtype=float32),
             'natural_language_embedding': Tensor(shape=(512,), dtype=float32),
             'natural_language_instruction': string,
-            'robot_state': Tensor(shape=(15,), dtype=float32),
+            'robot_state': Tensor(shape=(15,), dtype=float32, description=Explanation of the robot state can be found at https://sites.google.com/corp/view/berkeley-ur5),
         }),
         'reward': Scalar(shape=(), dtype=float32),
     }),

docs/catalog/berkeley_cable_routing.md

@@ -47,9 +47,9 @@ Split     | Examples
 FeaturesDict({
     'steps': Dataset({
         'action': FeaturesDict({
-            'rotation_delta': Tensor(shape=(3,), dtype=float32),
+            'rotation_delta': Tensor(shape=(3,), dtype=float32, description=Angular velocity about the z axis.),
             'terminate_episode': float32,
-            'world_vector': Tensor(shape=(3,), dtype=float32),
+            'world_vector': Tensor(shape=(3,), dtype=float32, description=Velocity in XYZ.),
         }),
         'is_first': bool,
         'is_last': bool,

docs/catalog/berkeley_fanuc_manipulation.md

@@ -48,20 +48,20 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(6,), dtype=float32),
-        'discount': Scalar(shape=(), dtype=float32),
+        'action': Tensor(shape=(6,), dtype=float32, description=Robot action, consists of [dx, dy, dz] and [droll, dpitch, dyaw]),
+        'discount': Scalar(shape=(), dtype=float32, description=Discount if provided, default to 1.),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'end_effector_state': Tensor(shape=(7,), dtype=float32),
-            'image': Image(shape=(224, 224, 3), dtype=uint8),
-            'state': Tensor(shape=(13,), dtype=float32),
-            'wrist_image': Image(shape=(224, 224, 3), dtype=uint8),
+            'end_effector_state': Tensor(shape=(7,), dtype=float32, description=Robot gripper end effector state, consists of [x, y, z] and 4x quaternion),
+            'image': Image(shape=(224, 224, 3), dtype=uint8, description=Main camera RGB observation.),
+            'state': Tensor(shape=(13,), dtype=float32, description=Robot joints state, consists of [6x robot joint angles, 1x gripper open status, 6x robot joint velocities].),
+            'wrist_image': Image(shape=(224, 224, 3), dtype=uint8, description=Wrist camera RGB observation.),
         }),
-        'reward': Scalar(shape=(), dtype=float32),
+        'reward': Scalar(shape=(), dtype=float32, description=Reward if provided, 1 on final step for demos.),
     }),
 })
 ```

docs/catalog/berkeley_gnm_cory_hall.md

@@ -48,21 +48,21 @@ FeaturesDict({
         'file_path': Text(shape=(), dtype=string),
     }),
     'steps': Dataset({
-        'action': Tensor(shape=(2,), dtype=float64),
-        'action_angle': Tensor(shape=(3,), dtype=float64),
-        'discount': Scalar(shape=(), dtype=float64),
+        'action': Tensor(shape=(2,), dtype=float64, description=Robot action, consists of 2x position),
+        'action_angle': Tensor(shape=(3,), dtype=float64, description=Robot action, consists of 2x position, 1x yaw),
+        'discount': Scalar(shape=(), dtype=float64, description=Discount if provided, default to 1.),
         'is_first': bool,
         'is_last': bool,
         'is_terminal': bool,
-        'language_embedding': Tensor(shape=(512,), dtype=float32),
+        'language_embedding': Tensor(shape=(512,), dtype=float32, description=Kona language embedding. See https://tfhub.dev/google/universal-sentence-encoder-large/5),
         'language_instruction': Text(shape=(), dtype=string),
         'observation': FeaturesDict({
-            'image': Image(shape=(64, 85, 3), dtype=uint8),
-            'position': Tensor(shape=(2,), dtype=float64),
-            'state': Tensor(shape=(3,), dtype=float64),
-            'yaw': Tensor(shape=(1,), dtype=float64),
+            'image': Image(shape=(64, 85, 3), dtype=uint8, description=Main camera RGB observation.),
+            'position': Tensor(shape=(2,), dtype=float64, description=Robot position),
+            'state': Tensor(shape=(3,), dtype=float64, description=Robot state, consists of [2x position, 1x yaw]),
+            'yaw': Tensor(shape=(1,), dtype=float64, description=Robot yaw),
         }),
-        'reward': Scalar(shape=(), dtype=float64),
+        'reward': Scalar(shape=(), dtype=float64, description=Reward if provided, 1 on final step for demos.),
     }),
 })
 ```