Torque control

@control

Command constraints

Necessary conditions for direct torque control

Torque commands must stay within limits and be continuous:

$$\begin{aligned} -\dot{\tau}_{j \max } & < \dot{\tau}_{jc} < \dot{\tau}_{j \max } \\ -\tau_{j \max } & < \tau_{jc} < \tau_{j \max } \end{aligned}$$

Direct torque control limits

xMateER3 Pro / xMateER7 Pro

Parameter	Axis 1	Axis 2	Axis 3	Axis 4	Axis 5	Axis 6	Axis 7	Unit
Max torque	85	85	85	85	36	36	36	$Nm$
Max torque derivative	1500	1500	1500	1500	1000	1000	1000	$Nm/s$

xMate3 / xMate7

Parameter	Axis 1	Axis 2	Axis 3	Axis 4	Axis 5	Axis 6	Unit
Max torque	85	85	85	36	36	36	$Nm$
Max torque derivative	1500	1500	1500	1000	1000	1000	$Nm/s$

Torque-related State Parameters

• These state data can be obtained in real-time mode. You can subscribe to the data by calling the startReceiveRobotState() interface, and then fetch their real-time values in the control loop using the getStateData() interface.

Data Name (fieldName)	Type	Unit	Meaning	Additional Notes
tau_m	ArrayXD	Nm	Joint torque / Link-side sensor torque feedback	Pure feedback from the sensor, does not include friction.
tau_c	ArrayXD	Nm	Commanded joint torque	SDK sent command value. Gravity compensation and other parameters need to be added to become the final torque command value in the system. Valid only after real-time mode is enabled.
tau_filtered_m	ArrayXD	Nm	Filtered joint torque	The filtered result of the link-side feedback joint torque tau_m. Valid only after real-time mode is enabled.
tau_vel_c	ArrayXD	Nm/s	Commanded torque derivative	The derivative of tau_c. Valid only after real-time mode is enabled.
tau_ext_base	Array6D	Nm	External torque in base frame	Valid only after real-time mode is enabled.
tau_ext_stiff	Array6D	Nm	External torque in stiffness frame	Valid only after real-time mode is enabled.
theta_m	ArrayXD	rad	Motor position	Position feedback from the motor side.
theta_vel_m	ArrayXD	rad/s	Motor velocity	Velocity feedback from the motor side.
motor_tau	ArrayXD	Nm	Motor torque	The actual torque feedback from the motor, includes friction. This feedback value can be converted into the joint feedback torque (i.e., link-side torque converted via motor force) by multiplying it with the reduction ratio.
motor_tau_filtered	ArrayXD	Nm	Filtered motor torque	The filtered result of the motor feedback torque, which is also the final sent command. You can use setFilterFrequency to configure the filter parameters. Valid only after real-time mode is enabled.

4. Common issues

Sag under direct torque control

Torque behavior depends on arm hardware, load parameters, and force-control model accuracy.

• Verification (pick one)

  • Enable drag mode, hold the drag button, and check whether the arm stays still (no float or sag).
  • Run RL force-control related routines and confirm they behave normally.

If either check fails, there is likely a force-control model error.

• Recommendations

  • If the model is wrong:
    • Set load mass and COM accurately (do not guess).
    • Run force-sensor calibration via the HMI interface or the SDK's calibrateForceSensor interface. When controlling the robot's motion through the SDK, it is recommended to call the calibrateForceSensor interface in the program to calibrate the force sensor before each movement.
  • If the model is fine:
    • Check commanded torques first.
    • Reason for robot drifting under zero-torque control: During torque control, the controller's underlying layer compensates for friction and gravity, so the actual commanded torque executed by the servo is friction + gravity + commanded torque sent by SDK. Since static friction lacks a precise mathematical model, inadequate torque compensation may cause the robot to move; this can be addressed by adjusting the friction compensation coefficient.

• Debugging tips

  • Start with small magnitude and low slew rate, then ramp up.
  • Avoid blocking or heavy work in the control callback to keep the torque loop stable.
  • If network jitter is present, prefer a direct, simple link.