Skip to main content
Common device parameters include identity parameters (serial number, firmware version, handedness) and network parameters (IP address, data port). Every device connected through the Wuji SDK (Wuji Hand 2, Wuji Glove, and more) accesses them through the same resource-style API. This page covers four cross-device mechanisms: parameter read/write, SDK user management with data import/export, local calibration storage with per-user isolation, and resource introspection. For each product’s parameter list and device operations, see the product docs linked in each section.

Reading Parameters

Read device properties through the semantic API:

Writing Parameters

Some parameters support writing:
The IP must be unique on the subnet. Avoid addresses already used by other devices or the host network adapter.
The IP must be unique on the subnet. Avoid addresses already used by other devices or the host network adapter.
Changing network parameters (IP address, port) may require reconnecting to the device.

Device Operations

Parameter persistence, reboot, and other device operations vary by product:

SDK User Management

When several operators share the same device, calibration and local parameters from one operator overwrite another’s. The SDK keeps a local user profile per operator at ~/.wuji/sdk/users/<user_id>/, isolating each user’s parameter and calibration files. Without a named user, parameter files and tactile calibration land in ~/.wuji/sdk/params/ and ~/.wuji/sdk/tactile/ directly, and pose calibration isn’t available at all. Create a named user when you need to isolate by operator, or when you need pose calibration:
Switching reloads calibration parameters for connected devices immediately, no reconnect needed. Deleting the current user automatically switches back to the default user. For measured edges around creating, updating, switching, and deleting users, and how the change persists across processes and languages, see Local SDK User Management.

User Data Export and Import

Pack everything an SDK user owns into a single zip file to back it up or move it to another host. A bundle holds the user’s hand calibration models, the tactile model for each device, and the latest complete tactile calibration run.
In Python, all three methods return a report dict whose components list the state of each component: Import behavior:
  • The owner recorded in the bundle is created if it doesn’t exist, and the current user stays the same
  • Every present component that isn’t skipped is written and overwrites that user’s existing file
  • Skipping wins over writing: a skipped component is never written and never overwrites an existing file, and its state in the report is skipped. Hand models are skipped this way when the owner is the default user (which doesn’t load local hand calibration), while the other components are still written
  • Validation (path safety, size caps, sha256, data format, and owner id) finishes before anything is written. Any failure raises an exception and leaves no file behind
The export path must end in .zip, and the SDK creates the parent directory if it’s missing. The three C calls return only a status code, because the per-component detail doesn’t cross the FFI boundary — use the Python API when you need per-component results. A missing bundle returns WUJI_STATUS_ERR_NOT_FOUND, and a bundle that fails validation returns WUJI_STATUS_ERR_INVALID_DATA.

Tactile Data Paths

The tactile model is trained in Wuji CLI tactile contact calibration. Models and calibration data are scoped by SDK user and device serial number: model/ holds the trained tactile model (contact.safetensors and contact.npz), and calibration-data/ holds the raw calibration runs, one subdirectory per run. A complete run has one data file per guided motion, 4 in total. Export takes the latest complete run only.

Calibration

Calibration fits the SDK’s algorithms to a specific physical device. The results are written to SDK-local parameter and model files and loaded automatically when the device subscribes to them. Calibration files are scoped by the current SDK user first, then split into three resource types: Parameter files and the tactile directory are still keyed by device serial number. The tactile directory holds the trained model under model/ and the latest calibration run under calibration-data/. The pose hand URDF is stored per SDK user and hand side only (such as left_hand.urdf / right_hand.urdf), without the serial number or hand-profile dimension, so any device of the same side under the same user shares one model. The default user has no pose hand model at all: calibration and import never write one, and real-time IK always falls back to the built-in default. For Wuji Glove specifics, see Wuji Glove Calibration. Switching the SDK user reloads the matching files for connected devices automatically, no reconnect required.
Calibration is limited under the default user. When multiple operators share a device, switch to a named user first:
  • Running calibrate as the default user fails immediately with ExecuteError(operation="calibrate", code=0x4108). Call SdkManager.create_user(...) and switch_user(...) to move to a named user before calibrating, so the generated URDFs are scoped to that user.
  • Setting a custom hand URDF as the default user fails immediately: the semantic API glove.hand_model_path().set(path) and the generic write paths (device.set, manager.set) all reject writes to calibration.hand_model_path with UnsupportedOperation and leave the parameter store untouched. Switch to a named user first. The semantic API also validates the path and rejects an unreadable file with InvalidConfig without saving it.
  • Under the default user, IK-derived streams (hand_skeleton, hand_joint_angles, tip_poses) always use the SDK’s built-in default URDF and ignore any stored calibration.hand_model_path. This prevents accidentally reusing another operator’s calibration when the SDK user wasn’t explicitly switched.
  • When to calibrate: first time you use a device, a different wearer puts on a wearable device, higher-level model or algorithm generation changes, or calibration accuracy has visibly degraded. The exact triggers depend on the product
  • Failure behavior: when calibration fails, the SDK preserves the last successful calibration and rolls back the new files automatically, so you can retry directly. Typical failure modes: timeout, solver fails to converge, file persistence fails, or the SDK user switches mid-calibration. Exact exception types vary by product
  • Backup and migration: use user data export and import to pack a whole user’s calibration results — on import the SDK rebuilds the paths under the target host’s user directory. Copying the directory works too, but the parameter file can store absolute paths pointing to other files under ~/.wuji/sdk/ (such as model files). When migrating to a host with a different home directory, update those paths or run calibration again
For each product’s calibration procedure, see the product docs:

Available Parameters

Listing All Resources

List all available resources, parameters, and topics on a device:

Interface List

Every C string getter uses a two-step size query: pass buf = NULL and buf_len = 0 to learn the required byte count, then allocate and call again to fill the buffer. APIs that return a report dictionary in Python return only a WujiStatus in C, because the per-component detail doesn’t cross the FFI boundary. For the WujiUserInfo and WujiUserUpdate fields, memory ownership, and the release functions, see the C Interface Reference.