Pose accuracy problem of the Franka Research 3 Robot

[JonesLLL]

Hi, our lab are currently working on a high-precision task on Franka Research 3 robot. During testing, we observed discrepancies between the actual flange position and the pose reading from robot_state.O_T_EE.

To verify the readings, I conducted two simple tests:

1. X-direction test: Placed the flange flat on an aluminum profile surface exactly 414.5 mm from the base in the X-direction.
2. Z-direction test: Placed the flange on an optical plate exactly 20 mm below the base in the Z-direction.

In the first case, the reported X-position was 418 mm (Fig. 1), which is reasonably close to the expected value. However, in the second case, the reported Z-position was -10 mm (Fig. 2) instead of the expected -20 mm.

What would be potential causes for this discrepancy and possible solution?

Figure 1. Position test in x-direction Figure 2. Position test in z-direction

[Siva552]

Maybe you can ask this question on Franka's community. Here is the website:https://www.franka-community.de. I also curious about this question. Can you share the result here if you get some reply?

[winkma]

Hi there!
The FR3 is pretty good in position repeatibility, which is <+/- 0.1 mm evaluated according to ISO 9283 standard. That's also what is listed in our datasheet. Position accuracy is generally worse than that, however we do not evaluate it across a large number of robots, so we don't specify it in the datasheets. The 10 mm difference that you have reported, I've also measured in our lab before with some FR3s, so that's not unusual.
What I would recommend for you if you need to reach positions with a low error:

1. try to pre-teach the pose, so you can utilize the 0.1 mm repeatibility as often as possible
2. if that doesn't work for your use case, try to use external position sensors (e.g. cameras, laser trackers) to correct your robots motions
3. This one is a bit more work, but there's a few cool papers (see examples below) that calibrate/identify the kinematics of Franka robots in the field - maybe check those out.

I'd love to hear if any of those work for you and probably the community would be thankful too since this is a challenge for many people 😊 We're also working on methods to bring the overall accuracy of our robots up and we'll communicate through the official channels once we have new features available for you.

Best,
Marius

[1] G. Franzese, M. Spahn, J. Kober, and C. D. Santina, “MUKCa: a Simple yet Effective Calibration Method to Democratize Calibration of Cobots”.
[2] “Philippe Nadeau | Robot Arm Kinematic Calibration.” Accessed: Mar. 17, 2025. [Online]. Available: http://philna.de/au-is/projects/robot-calibration/