Non-zero values from O_F_ext_hat_K without applied external force

[bdlim99]

Description:
I'm observing non-zero values from the O_F_ext_hat_K topic even when no external force is applied to the robot's end effector (EE) and no accessories are attached.

Steps to Reproduce:

1. Launch the Franka controller using:

   roslaunch franka_control franka_control.launch
   

2. In another terminal, run:

   rostopic echo franka_state_controller/franka_states | grep O_F_ext
   

3. Observe output similar to:

   O_F_ext_hat_K: [1.4046984925349721, -0.3746988189998095, 1.7208226415931065, 0.0025019231454423, 0.4513628440880607, -0.1473264999079312]
   O_F_ext_hat_K: [1.3852304678927174, -0.36299994795652923, 1.7134695277137726, -0.006568094169452518, 0.4395951988301791, -0.14457598724463322]
   O_F_ext_hat_K: [1.3910224549444952, -0.36906682006852587, 1.7129011732986914, -0.002232087659523152, 0.4441428239831187, -0.14767820703593915]
   ...
   

Observations:

• With the robot held in a fixed pose, the O_F_ext_hat_K value remains constant, suggesting that these readings are not merely random noise.
• When changing the robot's pose and then removing any external forces, the reported O_F_ext_hat_K values change accordingly, indicating a dependency on the robot's configuration.

Questions:

1. Is O_F_ext_hat_K intended to directly represent the external wrench acting on the robot's EE? If so, why are the values non-zero when no external force is applied?
2. Why does the value of O_F_ext_hat_K vary with the robot's pose, even in the absence of external disturbances?

Any clarification on the calibration process or the underlying estimation method that might explain this behavior would be greatly appreciated.

[winkma]

Hi @bdlim99 ,

The estimated external wrench O_F_ext_hat_K signal is calculated from the joint torque sensor measurements tau_J , using the robot model to compute the estimated external joint torques tau_ext_hat_filtered. Although mechanical wear can be a factor (e.g. friction), the biggest measurement errors here come from "torque sensor offsets". Those were calibrated in our production line just before you received the robot, but they are known to change over time, especially when there have been impacts to the torque sensors.
From the tau_ext_hat_filtered you can use the kinematics (aka the Jacobian) to compute O_F_ext_hat_K. That could be the reason why the error changes over time as in some configurations those joint torque offsets could be amplified or joint offsets in different joints could also cancel each other out.

I hope this was at least helpful to understand the issue, even though we cannot fix it for you right now 😒

If you have a Franka Research 3, please stay tuned for updates, as we will release a feature in Desk for users to re-calibrate the offsets! 😸

[shr-eyas]

@winkma Is that method to re-calibrate been introduced yet?
We have FR3 and would love to try it out!

[winkma]

Hey @shr-eyas , thanks for following up 😊 Yes, the torque calibration feature has been released with system image 5.8.0 🥳 Please try it out, and we'd love to hear your feedback on it if you have any!