DexForce · yhnsu · Jul 28, 2026 · Jul 29, 2026 · Jul 29, 2026 · Jul 29, 2026
diff --git a/agent_context/topics/robot-system/robot-system.md b/agent_context/topics/robot-system/robot-system.md
@@ -33,7 +33,7 @@ ObjectBaseCfg          uid, init_pos, init_rot, init_local_pose
   │                     disable_self_collision, init_qpos, body_scale,
   │                     build_pk_chain, use_usd_properties
       └─ RobotCfg      control_parts, urdf_cfg, solver_cfg, drive_pros (override default to "force")
-          ├─ DexforceW1Cfg   version, arm_kind, with_default_eef
+          ├─ DexforceW1Cfg   version, with_default_eef
           └─ CobotMagicCfg   (dual-arm defaults)
 ```
 
@@ -46,7 +46,7 @@ Key fields on `RobotCfg`:
 | `solver_cfg` | `SolverCfg \| Dict[str, SolverCfg] \| None` | IK solver config; dict keys must match `control_parts` keys |
 | `drive_pros` | `JointDrivePropertiesCfg` | Default drive type is `"force"` (overrides Articulation's `"none"`) |
 | `attrs` | `RigidBodyAttributesCfg` | Rigid-body physics attributes (mass, friction, damping, ...) |
-| variant fields | `enum \| str \| bool` | Optional subclass fields (e.g. `version`, `arm_kind`, `with_default_eef`) |
+| variant fields | `enum \| str \| bool` | Optional subclass fields (e.g. `version`, `with_default_eef`) |
 | `_pk_urdf_path` | `property \| method → str` | URDF for the FK/IK serial chain (one source, so it can't drift from sim) |
 
 ## The robot config protocol
@@ -134,7 +134,7 @@ Serialization (`to_dict` / `save_to_file`) is inherited — no need to implement
 
 | Robot | Config Class | Module | Structure | Notes |
 |---|---|---|---|---|
-| DexForce W1 | `DexforceW1Cfg` | `embodichain/lab/sim/robots/dexforce_w1/` | Package (`cfg.py`, `types.py`, `params.py`, `utils.py`) | Humanoid; versions: V021; arm kinds: ANTHROPOMORPHIC, INDUSTRIAL; component types: chassis, torso, eyes, head, left/right arm/hand |
+| DexForce W1 | `DexforceW1Cfg` | `embodichain/lab/sim/robots/dexforce_w1/` | Package (`cfg.py`, `types.py`, `params.py`, `utils.py`) | Humanoid; versioned chassis, torso, head, left/right arm and hand components |
 | CobotMagic | `CobotMagicCfg` | `embodichain/lab/sim/robots/cobotmagic.py` | Single file | Dual-arm; 6-DOF arms + 2-DOF grippers; uses OPW solver |
 
 ## Common Failure Modes

diff --git a/...tic/robots/dexforcew1_anthropomorphic.jpg → docs/source/_static/robots/dexforcew1.jpg b/...tic/robots/dexforcew1_anthropomorphic.jpg → docs/source/_static/robots/dexforcew1.jpg
diff --git a/docs/source/_static/robots/dexforcew1_industrial.jpg b/docs/source/_static/robots/dexforcew1_industrial.jpg
diff --git a/docs/source/features/workspace_analyzer/workspace_analyzer.md b/docs/source/features/workspace_analyzer/workspace_analyzer.md
@@ -33,8 +33,7 @@ sim = SimulationManager(SimulationManagerCfg(headless=False, sim_device="cpu"))
 # Add robot
 robot = sim.add_robot(DexforceW1Cfg.from_dict({
     "uid": "dexforce_w1", 
-    "version": "v021", 
-    "arm_kind": "industrial"
+    "version": "v021"
 }))
 
 # Quick analysis with defaults
@@ -175,7 +174,7 @@ sim = SimulationManager(SimulationManagerCfg(headless=False, sim_device="cpu"))
 
 # Add robot
 robot = sim.add_robot(DexforceW1Cfg.from_dict({
-    "uid": "dexforce_w1", "version": "v021", "arm_kind": "industrial"
+    "uid": "dexforce_w1", "version": "v021"
 }))
 
 # 1. Joint Space Analysis

diff --git a/docs/source/guides/add_robot.rst b/docs/source/guides/add_robot.rst
@@ -73,7 +73,7 @@ Key parameters
 | ``attrs``           | RigidBodyAttributesCfg           | Rigid-body physics attributes    |
 +---------------------+----------------------------------+----------------------------------+
 | variant fields      | enum / str / bool                | Optional subclass fields         |
-|                     |                                  | (e.g. ``version``, ``arm_kind``) |
+|                     |                                  | (e.g. ``version``)               |
 +---------------------+----------------------------------+----------------------------------+
 | ``_pk_urdf_path``   | property or method → str         | URDF for the FK/IK serial chain  |
 +---------------------+----------------------------------+----------------------------------+

diff --git a/docs/source/overview/sim/solvers/srs_solver.md b/docs/source/overview/sim/solvers/srs_solver.md
@@ -31,7 +31,6 @@ SRSSolver is configured via the `SRSSolverCfg` class, allowing detailed control
 from embodichain.data import get_data_path
 from embodichain.lab.sim.robots.dexforce_w1.types import (
     DexforceW1ArmSide,
-    DexforceW1ArmKind,
     DexforceW1Version,
 )
 from embodichain.lab.sim.robots.dexforce_w1.params import (
@@ -41,7 +40,6 @@ from embodichain.lab.sim.solvers.srs_solver import SRSSolver, SRSSolverCfg
 
 arm_params = W1ArmKineParams(
     arm_side=DexforceW1ArmSide.RIGHT,
-    arm_kind=DexforceW1ArmKind.ANTHROPOMORPHIC,
     version=DexforceW1Version.V021,
 )
 

diff --git a/docs/source/resources/robot/dexforce_w1.md b/docs/source/resources/robot/dexforce_w1.md
@@ -1,33 +1,29 @@
 
 # Dexforce W1
 
-Dexforce W1 is a versatile robot developed by DexForce Technology Co., Ltd., supporting both industrial and anthropomorphic arm types. It is suitable for various simulation and real-world application scenarios.
+Dexforce W1 is a dual-arm humanoid robot developed by DexForce Technology Co., Ltd.
 
 <div style="display: flex; justify-content: center; align-items: flex-start; gap: 20px; flex-wrap: wrap;">
   <figure style="text-align: center; margin: 10px;">
-    <img src="../../_static/robots/dexforcew1_anthropomorphic.jpg" alt="Anthropomorphic Version" style="height: 400px; width: auto;"/>
-    <figcaption><b>Anthropomorphic Version</b></figcaption>
-  </figure>
-  <figure style="text-align: center; margin: 10px;">
-    <img src="../../_static/robots/dexforcew1_industrial.jpg" alt="Industrial Version" style="height: 400px; width: auto;"/>
-    <figcaption><b>Industrial Version</b></figcaption>
+    <img src="../../_static/robots/dexforcew1.jpg" alt="Dexforce W1" style="height: 400px; width: auto;"/>
+    <figcaption><b>Dexforce W1</b></figcaption>
   </figure>
 </div>
 
 ## Key Features
 
-- Supports multiple arm types (industrial, anthropomorphic)
+- Supports dual 7-DOF arms
+- Supports version-owned asset layouts and calibration parameters
 - Configurable left/right hand brand and version
 - Flexible URDF assembly and simulation configuration
 - Compatible with SimulationManager simulation environment
 
 ## Method 1: Fine-grained configuration with `build_dexforce_w1_cfg`
 
-This method allows you to specify detailed parameters for each arm and hand. Recommended for advanced users who need full control over robot hardware options.
+This method allows you to specify detailed parameters for each arm and hand.
 
 **Parameters:**
 
-- `arm_kind`: Arm type, e.g., `DexforceW1ArmKind.ANTHROPOMORPHIC` or `DexforceW1ArmKind.INDUSTRIAL`.
 - `hand_types`: Dict specifying hand brand for each arm side (`LEFT`/`RIGHT`).
 - `hand_versions`: Dict specifying hand version for each arm side.
 
@@ -37,11 +33,11 @@ hand_types = {
     DexforceW1ArmSide.RIGHT: DexforceW1HandBrand.BRAINCO_HAND,
 }
 hand_versions = {
-    DexforceW1ArmSide.LEFT: DexforceW1Version.V021,
-    DexforceW1ArmSide.RIGHT: DexforceW1Version.V021,
+    DexforceW1ArmSide.LEFT: DexforceW1Version.V025,
+    DexforceW1ArmSide.RIGHT: DexforceW1Version.V025,
 }
 cfg = build_dexforce_w1_cfg(
-    arm_kind=DexforceW1ArmKind.ANTHROPOMORPHIC,
+    version=DexforceW1Version.V025,
     hand_types=hand_types,
     hand_versions=hand_versions,
 )
@@ -56,14 +52,11 @@ This method allows fast setup using a dictionary, suitable for simple scenarios
 **Parameters:**
 
 - `uid`: Unique robot identifier (string).
-- `version`: Robot version, e.g., `v021`.
-- `arm_kind`: Arm type, e.g., `anthropomorphic` or `industrial` (string).
+- `version`: Robot version, e.g., `v021`, `v022`, or `v025`.
 
 ```python
 from embodichain.lab.sim.robots import DexforceW1Cfg
-cfg = DexforceW1Cfg.from_dict(
-    {"uid": "dexforce_w1", "version": "v021", "arm_kind": "anthropomorphic"}
-)
+cfg = DexforceW1Cfg.from_dict({"uid": "dexforce_w1", "version": "v025"})
 robot = sim.add_robot(cfg=cfg)
 print("DexforceW1 robot added to the simulation.")
 ```
@@ -98,7 +91,94 @@ Choose `build_dexforce_w1_cfg` for maximum flexibility and hardware customizatio
 
 | Type                    | Options / Values                                      | Description                        |
 |-------------------------|-------------------------------------------------------|------------------------------------|
-| `DexforceW1ArmKind`     | `ANTHROPOMORPHIC`, `INDUSTRIAL`                       | Arm type                           |
 | `DexforceW1HandBrand`   | `BRAINCO_HAND`, `DH_PGC_GRIPPER`, `DH_PGC_GRIPPER_M`  | Hand brand                         |
-| `DexforceW1Version`     | `V021`                                                | Release version                    |
+| `DexforceW1Version`     | `V021`, `V022`, `V025`                                | Release version                    |
 | `DexforceW1ArmSide`     | `LEFT`, `RIGHT`                                       | Left/right hand identifier         |
+
+## Unified asset layout and version extension
+
+V022 and V025 use one unified Hugging Face archive per release:
+
+```text
+dexforce_w1/<version>/w1.zip
+└── w1/
+    ├── robot.urdf
+    ├── chassis.urdf
+    ├── torso.urdf
+    ├── head.urdf
+    ├── left_arm.urdf
+    ├── right_arm.urdf
+    ├── visual/
+    └── collision/
+```
+
+The runtime downloads each release archive once. Direct FK/IK uses `robot.urdf`
+or the arm URDFs, while configurable robot assembly reads all components from
+the same extracted directory. V022 and V025 assets are resolved through the
+registered Hugging Face dataset archives and the shared asset cache.
+
+### Version-owned end-effector offset
+
+Different arm revisions may place the physical mounting surface at different
+positions relative to the arm `ee` frame. This difference belongs to the W1
+revision, not to a BrainCo hand, DH gripper, PIKA gripper, or any other
+end-effector.
+
+`W1VersionSpec.default_eef_attach_xpos` is the single source of truth for this
+revision offset:
+
+| Version | Left arm | Right arm |
+|---------|----------|-----------|
+| V021 | Identity | Identity |
+| V022 | Identity (provisional; calibration required) | Identity (provisional; calibration required) |
+| V025 | `+0.012 m` along the `ee` frame Z axis | `+0.012 m` along the `ee` frame Z axis |
+
+The final assembly transform and solver TCP are derived as follows:
+
+```python
+final_attach_xpos = version_attach_xpos @ eef_attach_xpos
+final_tcp = version_attach_xpos @ solver_tcp
+```
+
+Therefore, the V025 offset affects both the assembled end-effector position and
+FK/IK results. The offset is also applied when callers provide a custom
+`hand_attach_xposes` value, a custom `left_hand`/`right_hand` component
+transform, or an explicit arm TCP through `DexforceW1Cfg.from_dict`.
+Serialization removes the derived offset and restores it on loading, so a
+configuration round trip does not apply the offset twice.
+
+Do not manually add the 12 mm correction to an end-effector transform or TCP.
+Those values must describe the end-effector relative to the standard mounting
+surface; the W1 version layer adds the robot revision correction.
+
+Per-component version overrides are supported for controlled migrations:
+
+```python
+cfg = DexforceW1Cfg.from_dict(
+    {
+        "uid": "dexforce_w1",
+        "version": "v021",
+        "component_versions": {
+            "left_arm": "v025",
+            "right_arm": "v025",
+        },
+    }
+)
+```
+
+To add another W1 revision:
+
+1. Register the new `DexforceW1Version` value and dataset archive.
+2. Add one `W1VersionSpec` entry in
+   `embodichain/lab/sim/robots/dexforce_w1/specs.py`.
+3. Register component URDF paths and the full-robot URDF path.
+4. Set the arm kinematic parameters and
+   `default_eef_attach_xpos` for both sides. Use identity only after confirming
+   that the arm `ee` frame is already at the physical mounting surface.
+5. Verify that the same version offset is present in both the assembled
+   end-effector pose and the final FK/IK TCP.
+6. Validate hand/gripper mounting, wrist cameras, FK/IK, VR teleoperation, and
+   real2sim task regression.
+
+Builders, control parts, assembly, TCP selection, analytical parameters and
+FK/IK then use the version specification without revision-specific branches.
diff --git a/docs/source/tutorial/modular_env.rst b/docs/source/tutorial/modular_env.rst
@@ -105,7 +105,6 @@ The main environment configuration inherits from :class:`envs.EmbodiedEnvCfg` an
 Uses the pre-configured :class:`DexforceW1Cfg` with customizations:
 
 - **Version**: Specific robot variant (v021)
-- **Arm Type**: Anthropomorphic configuration
 - **Position**: Initial placement in the scene
 
 **Sensor Configuration**
@@ -241,4 +240,4 @@ This tutorial demonstrates the full power of EmbodiChain's modular environment s
    **Using an AI coding agent?** These skills can help you build on this tutorial:
 
    - **/add-task-env** — Scaffold a new task environment with the correct file structure, ``@register_env`` decorator, base class methods, ``__init__.py`` update, and test stub.
-   - **/add-functor** — Add observation, reward, event, or randomization functors with the correct signature and module placement.
+   - **/add-functor** — Add observation, reward, event, or randomization functors with the correct signature and module placement.