Software, SDK & teleoperation
Skye uses the same software as Luna: GentoPlatform for hands-on control, the Gento SDK for your own programs, and Apex for VR teleoperation. Get comfortable with GentoPlatform first; understanding how the robot behaves makes code development easier.
GentoPlatform
GentoPlatform (FX_Platform) is the host application for controlling and debugging Gento series robots. It runs on Linux x86 and Windows 64-bit, ships with the SDK, and the current version is for controller main version 1004. The interface is the same as on Luna; GentoPlatform walks through every panel.
- Get the software from the Gento series SDK repository. Executables are provided for Windows and Ubuntu 20.04 x86; on other systems, build from source.
- To run from source, install Python 3.10 or later, pyinstaller, and pillow, and make sure
SDK_PYTHONcontains the latestlibMarvinSDK.soandlibKine.so, compiled under./contrlSDKand./kinematicsSDK. Then runpython ui.py. To build an executable for a computer without Python, runpython setup.py. - Check that the controller is powered, the Ethernet cable runs from the controller port marked CAT.6 to your computer or a switch, and the E-stop on the robot is released.
- In the connection area at the top left, enter the motion controller's IP (
6.6.7.190) and click Connect Robot. When the connection succeeds, the button turns green and reads Connected.
If the connection fails, check that your computer and the controller are on the same LAN, check the IP address, then check the robot's power and network settings.
For first use and testing, set speed and acceleration to 5–10%.
Gento SDK
The motion control SDK wraps the low-level communication (L0) and provides interfaces for system management, motion control, state switching, parameter configuration, kinematics, and trajectory planning.
| Resource | Link |
|---|---|
| Gento series SDK repository | tianjizn/tianji-robot-SDK (Gento_Skye+Luna branch) |
| API documentation | tianjizn.github.io/api |
Directory structure
GENTO_SDK/
├── C_SDK/ # SDK source
│ ├── Common/ # Common headers
│ ├── FileClient/ # File transfer client
│ ├── Kinematics/ # ArmKinematics, BaseMath, DynaIdent, KineCommon,
│ │ # MotionPlanner, SkyeBodyKinematics
│ ├── L0Control/ # Low-level communication and control
│ └── L1Robot/ # Upper-level API
├── C_EXAMPLE/ # C++ examples that use C_SDK source directly
├── C_EXAMPLE_USE_DLL_SO/ # C++ examples that use the compiled DLL/SO
├── PYTHON_SDK/ # Python wrapper (GentoRobot.py)
├── PYTHON_EXAMPLE/ # Python examples
├── win_auto_compile.bat # One-step Windows build (source → DLL)
├── linux_auto_compile.sh # One-step Linux build (source → SO)
├── compile_methods.txt # Build commands for each platform
└── README.md
Version compatibility
Controller and SDK versions are MAJOR.MINOR.PATCH. The SDK connects only when MAJOR and MINOR match; otherwise the connection returns error -4, "Version incompatible". Call FX_L1_System_GetSDKVersion() (Python: get_sdk_version()) to get the SDK version, then upgrade or downgrade the controller to match.
Core modules
| Module | Description |
|---|---|
| System | Connect to and disconnect from the controller, set the log level, reboot, update firmware, transfer files |
| State machine | Switch between position, joint/Cartesian/force impedance, drag teaching (joint, X, Y, Z, rotation), and collaborative release |
| Real-time feedback | 1 kHz data (joint position, velocity, torque, IMU, F/T sensor) and 500 Hz slow-group data (configuration, diagnostics) |
| Parameters | Read and write integer, float, and string parameters by name |
| Terminal communication | Send and receive data with external devices over CAN FD or RS485 |
| Hardware configuration | Brake lock and release, encoder offset reset, soft-limit disable |
| Runtime motion | E-stop, joint position commands, force/torque control, scaling, stiffness and damping |
| Kinematics and planning | FK/IK, Jacobian, tool transforms, body kinematics, MoveJ/MoveL, multi-segment Cartesian, dual-arm synchronized planning |
| Dynamics identification | Identify payload mass, center of mass, and inertia from recorded data |
Core types
| Type | Description |
|---|---|
FXObjType | Robot component: left arm (FX_OBJ_ARM0), right arm (FX_OBJ_ARM1), head, torso, lifting platform |
FXStateType | Control state: idle, position, joint/Cartesian/force impedance, drag, release, error |
ROBOT_RT | Real-time feedback: joint position, velocity, torque, IMU, F/T sensor |
ROBOT_SG | Slow-group configuration and extended feedback |
FXFuncReturn | Return value: 0 is success; negative values are specific errors |
Usage notes
- SDK demo logic and parameters were written for R&D testing. Use them as a reference, not as production code.
- Stiffness and damping values are references and may change between controller versions; ask Gento support.
- The SDK uses UDP with a reliability protocol. Allow UDP on the configured ports (default 50000–50010) in your firewall.
FX_L1_System_Link()returns a positive latency in microseconds on success; negative values are errors.- Always call
FX_L1_System_Unlink()before your program exits, to release sockets and resources. - The README in the SDK covers quick start, building, and FAQs.
For the chassis, which uses ROS 2, see Chassis communication.
VR teleoperation
With the Apex teleoperation kit, an operator controls Skye from a VR headset with a real-time first-person view. The KM Data Converter toolchain turns recordings into imitation-learning datasets, for training and then inference on the real robot.
| Part | Description |
|---|---|
| Teleoperation kit | PICO Ultra headset, controllers, and three motion trackers, plus a GameSir motion gamepad. Software comes preinstalled |
| Front end | ROS-based host application for Ubuntu and Windows: connect to the back end, start the robot, switch modes, record and play back data. Install it on your computer; it is also preinstalled on the robot's AI computer |
| Back end | Runs on the Orin: ROS 2 nodes, motion control, cameras, and recording |
| VR headset | Meta Quest and Pico; the teleoperation app comes preinstalled |
| Options | Manus data gloves and Wuji Hand; a ROS 2 SDK for integrating with existing ROS 2 systems |
| Service | systemd unit | Description |
|---|---|---|
| Back-end API | apex-backend.service | HTTP/WebSocket entry point |
| Cameras | apex-camera.service | 4-channel GMSL cameras |
| Robot control | apex-robot.service | ROS control and state publishing |
| Teleoperation | apex-teleop.service | IK, planning, and command multiplexing |
For wiring, wearing the trackers, and first startup, see Skye and Luna: first teleoperation. To install or update the software, see Install or update Apex. The ROS 2 topics and services are listed in Skye and Luna ROS 2 interfaces.
After every power-up of the Orin or Thor, start the cameras before you teleoperate. To check the installed Apex version, see Check the Apex teleoperation version.
- Use a wired network where possible; teleoperating over the headset's Wi-Fi is not recommended.
- If the front end can't connect to the robot, check the power, network cables, router or switch, and that the IP subnets match.
- Before you start, check the E-stop, controller power, communication cables, USB drive, and headset battery.
- Switch to standby when you aren't teleoperating, to avoid accidental input.
- Back up the original configuration files before you change the IP, payload, stiffness, damping, Home point, or TCP offset.
- Stop immediately if the headset runs out of battery, loses power, or shuts down.
Next: maintenance & safety.