ROVIO test with Quadrotor!

[gajena]

Hi @ZacharyTaylor , Thanks for this great wiki!, I was trying this thing but having some issue!

Problem Description

I am trying to run nMPC with ROVIO on quadrotor. Followed the procedure given:

• System Identification with 70% fitting.
• Then Flight with VICON : Successful Test.
• Then Tested ROVIO with Manual Flight: ROVIO was working Great.
• Then started ROVIO based Flight and it starts to move around setpoint and also having twitching.
• Then tried with MSF on ROVIO+MAVIMU, but twitching has increased.

  System Config

  VIO

  • Pixracer with Bluefox MLC200W (hardware synced)
  • IMU at 100 Hz Using ethzasl_mav_px4 checkout on this commit
  • Camera at 20 hz

    MAV

  • Quadrotor X config, Ironman Tarrot
  • Pixhawk Autopilot

    Processing Unit

  • Intel NUC with 0cta-core processor

[ZacharyTaylor]

There are three possible issues that can result in this behavior.

1. a time delay in the system
2. MSF settings
3. IMU noise

1) is unlikely as ROVIO is very sensitive to this and wouldn't be working if it was out by much. 2) it could be that the given TF or some other setting is off, but this seems unlikely if it is only twitching a little but not diverging. 3) if it is IMU noise the easiest way to see this is to look at the velocity estimated by MSF. If the IMU has insufficient filtering then you will find that moving the system by hand results in smooth estimates, but as soon as the motors start you will get large and rapid random oscillations in the value. If this is the case you can a) add some foam or gel dampers around your PX4 b) decrease the frequency cutoff of the IMU filters or c) switch away from MSF to something less sensitive. (For example https://github.com/ethz-asl/odom_predictor is a much simpler approach that doesn't do any sensor fusion and doesn't try to reestimate the bias or velocity, which if ROVIO is working should already be accurate).

[SnShine]

Which nMPC are you using?

I've faced a similar issue in the past. If ROVIO is giving odom linear twist in global frame and your nMPC accepting odom linear twist in body frame you'll have the issue of drone making rounds around setpoint in some yaw angles (of setpoint) and will converge to setpoint as expected in some yaw angles of setpoint.

On a side note, I'm spent days making ROVIO work on EuroC dataset and gave up to use other repos. 😅

[gajena]

@SnShine I am using this repository mav_control_rw. I have already worked on ROVIO with high-end IMU like XSensMTi, but this time, I am trying with low-end IMUs. So, most probably it has large noise.

[gajena]

@ZacharyTaylor I have checked all the things and seems like cheaper IMUs has the problem with ROVIO. For now, I am running nMPC with RTK-GPS, and it is working great. Recently, we bought pixhawk2 Cube that has better IMUs with damping. So I will try with that and will let you know.