Crazyflie 2.1 Hardware Consistency Issue

[tomjz]

Dear Bitcraze,

We purchased two Crazyflie 2.1 in Feb 2021, and four more in Jan 2022.
For all of the quads, we use the firmware from https://github.com/bitcraze/crazyflie-firmware with the commit head de913f40a09792b7b41f3f37b3e75a2311c28b51.
I use the now deprecated ROS wrapper from https://github.com/whoenig/crazyflie_ros

With the same firmware and ROS wrapper, the two Crazyflies I got in 2021 is able to hover properly under Vicon using the the launch script hover_vicon.launch. However, the four new quads do not behave correctly using this set of code. Specifically, the new quads seem to jump up and down erratically. We have doubled checked the relevant ROS topics and they all seem to be correct. We also ran a hardware version check and there seems be no problems either. We also tried to fly the new quads using teleop with an xbox joystick and they worked ok, at least the operator did not feel anything wrong with the new quads.

We would like to know if there are some differences in these two versions of Crazyflie 2.1 which may have caused the issue. Please let us know if you need any other info. Appreciate your help!

Tom

[tobias]

Hi Tom,

That is definitely strange. There are no HW differences between the two batches and it sounds a bit like the IMU (BMI088) is performing differently. You say they are jumping, is it mainly Z that looks affected and all four new quads show the same behavior?

When flying manually the complimentary filter + PID is used which might be a bit more forgiving. What controller are you using?

Could the propellers from the new batch be more unbalanced? There is a simple propeller test in cfclient->console tab you could try.

There are logging variables for raw gyro and accelerometer (filtered) values. You could try and compare these.

[tomjz]

Thanks for the reply.

I assembled two of the new crazyflies and they behave similarly but I have yet to test the other two. It's not just the Z axis. It sometimes bounces back and forth in both x and y either, but Z is the most noticeable axis that has problems.

The controller that I am using is an Xbox joystick.

I did a propeller test using cfclient->console. The old (and working) quad gives the following:
HEALTH: Acc noise floor variance X+Y:0.004037, (Z:0.002166)

HEALTH: Motor M1 variance X+Y: 1.15 (Z:1.75), voltage sag:0.29
HEALTH: Motor M2 variance X+Y: 0.83 (Z:6.06), voltage sag:0.31
HEALTH: Motor M3 variance X+Y: 0.47 (Z:0.98), voltage sag:0.26
HEALTH: Motor M4 variance X+Y: 0.63 (Z:2.64), voltage sag:0.27

The new quad, however, gives

HEALTH: Acc noise floor variance X+Y:0.003090, (Z:0.002273)
HEALTH: Motor M1 variance X+Y: 0.72 (Z:1.73), voltage sag:0.16
HEALTH: Motor M2 variance X+Y: 1.72 (Z:7.03), voltage sag:0.15
HEALTH: Motor M3 variance X+Y: 0.49 (Z:2.27), voltage sag:0.16
HEALTH: Motor M4 variance X+Y: 1.82 (Z:2.54), voltage sag:0.16
STAB: WARNING: stabilizer loop rate is off (665)

Could the stabilizer loop warning be a problem? I will do more test and logging of the gyro values and acc to compare. Thanks again.

Tom

[tobias]

The propellers look balanced enough. The WARNING is due to the propeller test is hogging the loop and not an error. Are you running the same firmware as you are not getting the warning on the old crazyflie?

[tomjz]

The firmware are the same.

So I was finally able to use a geometric controller for these new quadcopters to do trajectory tracking, but with one difference:
The old quads have a trim motor command of 43000, but the new quads require a trim motor command of 53000. The minimum command also needs to change from 10000 to 20000 for the new quads to work.

I have not gotten the PID controller to work yet for hovering under motion capture system, but it may also come down to more tuning.
Would you suspect this difference to be a result of different propellers?
I will run more diagnostics later this week, and will update here if I have new findings.

[kimberly]

Hi!

Let me see if I can help out.

It could be that there is also differences in the NRF firmware as well, which has been done by the maintainers of Crazyswarm2. If you flash the newest firmware on the crazyflies through the cfclient, then least you'll know that the NRF has been updated to the latest.

It is a bit difficult to really determine what is wrong, but since you are using an old ROS package, is it perhaps an idea to try out Crazyswarm and see if you are seeing the same issues? In that case you will need to update the crazyradio firmware too.

[tobias]

If you have the possibility change the full power path between the new and old. That is battery, motors, props. Then check the hover thrust again. This would tell us if the any of these components are the issue.

[earny]

I'm experiencing the same behavior with a Crazyflie 2.1 bought this week. Seems like PID controller is not tuned for this hardware.

[kimberly]

@earny

Please start a discussion for this discussions.bitcraze.io, either under general or Q&A. We are moving all support over there.