Document TCP frames, hand joint mapping and tactile units

Clarifies observed TCP reference frames, axis directions, metre/radian units and rotation-vector representation; documents the 22-joint hand mapping and elastomer-frame tactile wrench units (N and N*m). Adds HOME-pose Isaac Sim coordinate illustrations and a machine-readable joint mapping. Camera diagrams show nominal URDF frames only. Recorded data, videos, annotations and calibration parameter packages are unchanged.

README.md

@@ -327,6 +327,30 @@ The 65D state, joint-torque, and action vectors use the following order:
 | 36-57 | `right_hand_j0` to `right_hand_j21` | Right dexterous-hand joints |
 | 58-64 | `motor_j0` to `motor_j6` | Torso / motor-related joints |
 
+#### Hand joint mapping to the public URDF
+
+The 22 hand joint positions are **absolute joint angles in radians (rad)**. The left hand occupies `observation.state[7:29]`, and the right hand occupies `observation.state[36:58]` (Python slices, end exclusive). The hand portions of `action` use the same joint order and angle unit.
+
+Use the following ordered suffixes with the `left_` or `right_` prefix to obtain the corresponding joint names in the [public Sharpa Wave URDF](https://github.com/sharpa-robotics/sharpa-urdf-usd-xml/tree/0d19cac602f46456b819e4b6a2c09a74982c9a3e/wave_01):
+
+```python
+hand_joint_suffixes = [
+    'thumb_CMC_FE', 'thumb_CMC_AA', 'thumb_MCP_FE', 'thumb_MCP_AA', 'thumb_IP',
+    'index_MCP_FE', 'index_MCP_AA', 'index_PIP', 'index_DIP',
+    'middle_MCP_FE', 'middle_MCP_AA', 'middle_PIP', 'middle_DIP',
+    'ring_MCP_FE', 'ring_MCP_AA', 'ring_PIP', 'ring_DIP',
+    'pinky_CMC', 'pinky_MCP_FE', 'pinky_MCP_AA', 'pinky_PIP', 'pinky_DIP',
+]
+left_joint_names = ["left_" + name for name in hand_joint_suffixes]
+right_joint_names = ["right_" + name for name in hand_joint_suffixes]
+left_hand_q = row["observation.state"][7:29]
+right_hand_q = row["observation.state"][36:58]
+```
+
+The nominal mapping is `q_urdf = q_dataset`. URDF `origin` defines the joint transform at zero angle, and `axis` defines the positive rotation direction. Do not subtract an episode's first frame as a new zero or apply an additional left/right sign flip. No per-device mechanical zero-offset correction is applied by this export.
+
+Two SDK naming aliases are worth distinguishing: `thumb_DIP` corresponds to URDF `thumb_IP`, and `pinky_CMC_FE` corresponds to URDF `pinky_CMC`. The tactile finger name `little` corresponds to URDF `pinky`. A complete index-to-name mapping for both hands is provided in [hand_joint_mapping.json](assets/coordinate_frames/hand_joint_mapping.json).
+
 ### TCP Pose State and Action
 
 `observation.state.tcp_pose` and `action.tcp_pose` are separate 12D float32 vectors. Components `0:6` contain the left-arm TCP pose and `6:12` contain the right-arm TCP pose.
@@ -335,7 +359,47 @@ The observation is read from the source `state/left_arm/tcp_pose` and `state/rig
 
 The existing 24D `observation.state.tcp` remains unchanged: left pose `0:6`, left force/torque `6:12`, right pose `12:18`, right force/torque `18:24`. Its two pose slices correspond to the new observed TCP-pose vector. The force/torque slices are 0 in every episode.
 
-The export does not fully declare TCP units, reference frames or rotation convention (`rotation_type` is null). Use the acquisition/controller convention before geometric transformations; do not infer Euler angles or quaternions from the vector width alone.
+`observation.state.tcp_pose` describes the left and right `hand_base_link` TCP frames relative to the robot base frame `root_link`. Each hand is represented as `[x, y, z, rx, ry, rz]`: position is in **metres (m)** and orientation is a **rotation vector in radians (rad)**, not Euler angles / roll-pitch-yaw or a quaternion. For a rotation vector `r`, its direction is the rotation axis and its norm is the rotation angle. Convert it with `scipy.spatial.transform.Rotation.from_rotvec(r)` or an equivalent Rodrigues implementation.
+
+The robot base uses a right-handed coordinate system: **+X forward, +Y toward the robot's left, and +Z upward**. Position values refer to the hand-base origin expressed in this base frame, not the camera frame or a fingertip contact point.
+
+The local TCP frames are `left_hand_base_link` and `right_hand_base_link`. In the URDF zero / straight-finger configuration, **+Z points from the wrist toward the fingers, +X points toward the palm / grasping side, and +Y points toward the little-finger side for the left hand and toward the thumb side for the right hand**. These local axes move with the hand. The columns of the decoded rotation matrix give the local TCP X, Y and Z directions expressed in `root_link`.
+
+The corresponding homogeneous transform is `T_root_hand_base`, with `p_root = T_root_hand_base @ p_hand_base` for homogeneous column vectors. This TCP frame is distinct from each fingertip's tactile `elastomer` frame. Existing metadata may still contain `rotation_type: null`; this README explicitly documents the observed TCP convention without rewriting the recorded arrays. The frame definition and agreement with nominal URDF FK do not establish independently measured physical positioning accuracy.
+
+#### Coordinate-frame illustrations
+
+The robot is shown in its configured **HOME pose**, rendered in Isaac Sim. The overview is a front view. Axis colors are **X red, Y green, Z blue**; left and right refer to the robot's sides. The illustrations explain coordinate definitions and do not establish physical calibration accuracy.
+
+![North in HOME pose: base, hand TCP and camera coordinate frames](assets/coordinate_frames/home_front.png)
+
+<details>
+<summary>Base and left/right TCP close-ups</summary>
+
+The two TCP origins are the corresponding `hand_base_link` origins. The local axes move with the hands; they are not fixed to the robot-base axes.
+
+![Robot base coordinate frame](assets/coordinate_frames/root_frame.png)
+
+![Left hand-base TCP coordinate frame](assets/coordinate_frames/left_tcp.png)
+
+![Right hand-base TCP coordinate frame](assets/coordinate_frames/right_tcp.png)
+
+</details>
+
+<details>
+<summary>Head and wrist camera coordinate-frame close-ups</summary>
+
+These camera diagrams use the **nominal URDF camera-link frames**. They are not recording-specific calibrated extrinsics. Camera intrinsics, distortion parameters and calibrated camera-to-hand transforms are outside this documentation update. Close-up viewing angles are chosen to make the 3D axes readable; the robot remains in the same HOME configuration.
+
+![Head-left camera coordinate frame](assets/coordinate_frames/head_left_camera.png)
+
+![Head-right camera coordinate frame](assets/coordinate_frames/head_right_camera.png)
+
+![Left wrist-camera coordinate frame](assets/coordinate_frames/left_wrist_camera.png)
+
+![Right wrist-camera coordinate frame](assets/coordinate_frames/right_wrist_camera.png)
+
+</details>
 
 ### Video Streams
 
@@ -448,6 +512,12 @@ The dataset provides tactile information in structured numeric and image-based f
 - `observation.images.tactile_deform` visualizes contact-induced deformation patterns.
 - `observation.images.tactile_raw` preserves the raw tactile camera layout for custom preprocessing and representation learning.
 
+The 60D numeric vector is ordered as **left hand first, then right hand**, with each hand ordered **thumb, index, middle, ring, little**. Each consecutive six-value block is `[Fx, Fy, Fz, Tx, Ty, Tz]` for one fingertip.
+
+Each fingertip's six-axis force/torque values are expressed in its corresponding **`*_elastomer` local coordinate frame** in the Sharpa Wave URDF. **Force components `Fx`, `Fy`, `Fz` are in newtons (N); torque components `Tx`, `Ty`, `Tz` are in newton-metres (N·m).** The export preserves the SDK F6 values without an additional coordinate rotation or unit scaling. Dataset `little` channels correspond to the URDF `pinky` finger.
+
+The `*_elastomer` frame is the tactile frame. The `*_fingertip` frame is a separate frame used for fingertip position / inverse-kinematics annotations, and `*_hand_base_link` is the hand-base TCP frame; these frames are not interchangeable.
+
 All tactile modalities are synchronized with visual observations, proprioception, and actions at 30 FPS. The 60D signal provides a compact tactile representation; the tactile video streams provide spatially resolved observations for models designed to process the additional input resolution.
 
 ## Usage Recommendations

assets/coordinate_frames/hand_joint_mapping.json

@@ -0,0 +1,194 @@
+{
+  "dataset": "Sharpa-Robotics/SharpaDex-v1.0",
+  "urdf_repo": "sharpa-robotics/sharpa-urdf-usd-xml",
+  "urdf_revision": "0d19cac602f46456b819e4b6a2c09a74982c9a3e",
+  "joint_angle_unit": "radian",
+  "mapping_convention": "Nominal direct joint-angle mapping, q_urdf=q_dataset. No per-device zero-offset correction is claimed. URDF origin specifies the q=0 transform; positive rotation follows its axis.",
+  "left_state_slice_python": [
+    7,
+    29
+  ],
+  "right_state_slice_python": [
+    36,
+    58
+  ],
+  "joints": [
+    {
+      "hand_index": 0,
+      "left_state_index": 7,
+      "right_state_index": 36,
+      "left_urdf_joint": "left_thumb_CMC_FE",
+      "right_urdf_joint": "right_thumb_CMC_FE",
+      "sdk_sample_name": "thumb_CMC_FE"
+    },
+    {
+      "hand_index": 1,
+      "left_state_index": 8,
+      "right_state_index": 37,
+      "left_urdf_joint": "left_thumb_CMC_AA",
+      "right_urdf_joint": "right_thumb_CMC_AA",
+      "sdk_sample_name": "thumb_CMC_AA"
+    },
+    {
+      "hand_index": 2,
+      "left_state_index": 9,
+      "right_state_index": 38,
+      "left_urdf_joint": "left_thumb_MCP_FE",
+      "right_urdf_joint": "right_thumb_MCP_FE",
+      "sdk_sample_name": "thumb_MCP_FE"
+    },
+    {
+      "hand_index": 3,
+      "left_state_index": 10,
+      "right_state_index": 39,
+      "left_urdf_joint": "left_thumb_MCP_AA",
+      "right_urdf_joint": "right_thumb_MCP_AA",
+      "sdk_sample_name": "thumb_MCP_AA"
+    },
+    {
+      "hand_index": 4,
+      "left_state_index": 11,
+      "right_state_index": 40,
+      "left_urdf_joint": "left_thumb_IP",
+      "right_urdf_joint": "right_thumb_IP",
+      "sdk_sample_name": "thumb_DIP"
+    },
+    {
+      "hand_index": 5,
+      "left_state_index": 12,
+      "right_state_index": 41,
+      "left_urdf_joint": "left_index_MCP_FE",
+      "right_urdf_joint": "right_index_MCP_FE",
+      "sdk_sample_name": "index_MCP_FE"
+    },
+    {
+      "hand_index": 6,
+      "left_state_index": 13,
+      "right_state_index": 42,
+      "left_urdf_joint": "left_index_MCP_AA",
+      "right_urdf_joint": "right_index_MCP_AA",
+      "sdk_sample_name": "index_MCP_AA"
+    },
+    {
+      "hand_index": 7,
+      "left_state_index": 14,
+      "right_state_index": 43,
+      "left_urdf_joint": "left_index_PIP",
+      "right_urdf_joint": "right_index_PIP",
+      "sdk_sample_name": "index_PIP"
+    },
+    {
+      "hand_index": 8,
+      "left_state_index": 15,
+      "right_state_index": 44,
+      "left_urdf_joint": "left_index_DIP",
+      "right_urdf_joint": "right_index_DIP",
+      "sdk_sample_name": "index_DIP"
+    },
+    {
+      "hand_index": 9,
+      "left_state_index": 16,
+      "right_state_index": 45,
+      "left_urdf_joint": "left_middle_MCP_FE",
+      "right_urdf_joint": "right_middle_MCP_FE",
+      "sdk_sample_name": "middle_MCP_FE"
+    },
+    {
+      "hand_index": 10,
+      "left_state_index": 17,
+      "right_state_index": 46,
+      "left_urdf_joint": "left_middle_MCP_AA",
+      "right_urdf_joint": "right_middle_MCP_AA",
+      "sdk_sample_name": "middle_MCP_AA"
+    },
+    {
+      "hand_index": 11,
+      "left_state_index": 18,
+      "right_state_index": 47,
+      "left_urdf_joint": "left_middle_PIP",
+      "right_urdf_joint": "right_middle_PIP",
+      "sdk_sample_name": "middle_PIP"
+    },
+    {
+      "hand_index": 12,
+      "left_state_index": 19,
+      "right_state_index": 48,
+      "left_urdf_joint": "left_middle_DIP",
+      "right_urdf_joint": "right_middle_DIP",
+      "sdk_sample_name": "middle_DIP"
+    },
+    {
+      "hand_index": 13,
+      "left_state_index": 20,
+      "right_state_index": 49,
+      "left_urdf_joint": "left_ring_MCP_FE",
+      "right_urdf_joint": "right_ring_MCP_FE",
+      "sdk_sample_name": "ring_MCP_FE"
+    },
+    {
+      "hand_index": 14,
+      "left_state_index": 21,
+      "right_state_index": 50,
+      "left_urdf_joint": "left_ring_MCP_AA",
+      "right_urdf_joint": "right_ring_MCP_AA",
+      "sdk_sample_name": "ring_MCP_AA"
+    },
+    {
+      "hand_index": 15,
+      "left_state_index": 22,
+      "right_state_index": 51,
+      "left_urdf_joint": "left_ring_PIP",
+      "right_urdf_joint": "right_ring_PIP",
+      "sdk_sample_name": "ring_PIP"
+    },
+    {
+      "hand_index": 16,
+      "left_state_index": 23,
+      "right_state_index": 52,
+      "left_urdf_joint": "left_ring_DIP",
+      "right_urdf_joint": "right_ring_DIP",
+      "sdk_sample_name": "ring_DIP"
+    },
+    {
+      "hand_index": 17,
+      "left_state_index": 24,
+      "right_state_index": 53,
+      "left_urdf_joint": "left_pinky_CMC",
+      "right_urdf_joint": "right_pinky_CMC",
+      "sdk_sample_name": "pinky_CMC_FE"
+    },
+    {
+      "hand_index": 18,
+      "left_state_index": 25,
+      "right_state_index": 54,
+      "left_urdf_joint": "left_pinky_MCP_FE",
+      "right_urdf_joint": "right_pinky_MCP_FE",
+      "sdk_sample_name": "pinky_MCP_FE"
+    },
+    {
+      "hand_index": 19,
+      "left_state_index": 26,
+      "right_state_index": 55,
+      "left_urdf_joint": "left_pinky_MCP_AA",
+      "right_urdf_joint": "right_pinky_MCP_AA",
+      "sdk_sample_name": "pinky_MCP_AA"
+    },
+    {
+      "hand_index": 20,
+      "left_state_index": 27,
+      "right_state_index": 56,
+      "left_urdf_joint": "left_pinky_PIP",
+      "right_urdf_joint": "right_pinky_PIP",
+      "sdk_sample_name": "pinky_PIP"
+    },
+    {
+      "hand_index": 21,
+      "left_state_index": 28,
+      "right_state_index": 57,
+      "left_urdf_joint": "left_pinky_DIP",
+      "right_urdf_joint": "right_pinky_DIP",
+      "sdk_sample_name": "pinky_DIP"
+    }
+  ],
+  "source_dataset_revision": "5c3715ace9a09a23fc79e7d21e75b4c06e19e859"
+}