Vision position estimation fusing with MSF framework, imu-camera setting problem

1

Hello guys!
I am testing vision position estimate in LPE mode.
I am using MSF framework to fuse vision and imu data.
I can get pretty good position estimation information from VISION_POSITION_ESTIMATE message(not using vision for heading, but it seems that is not enough, because when I switch to POSCTL mode, QuadCopter keep fly around in a circle. It looks like this:

07_00_06_0.png1024×768 46 KB

And this is my flying log file:http://logs.uaventure.com/view/QcABiUm82uMMLwzAU8SjBX

As mhkabir suggested, I should use vision for heading, I know by set ATT_EXT_HDG_M can enable this function. But I don’t really know how to mess with MSF settings. This is my MSF settings:

data_playback: false
core/core_fixed_bias: false

#########IMU PARAMETERS#######
####### pixhawk - MPU6050
core/core_noise_acc: 0.013
core/core_noise_accbias: 0.000108
core/core_noise_gyr: 0.00065
core/core_noise_gyrbias: 0.00000212

pose_sensor/pose_fixed_scale: false
pose_sensor/pose_noise_scale: 0.0
pose_sensor/pose_noise_p_wv: 0.0
pose_sensor/pose_noise_q_wv: 0.0
pose_sensor/pose_noise_q_ic: 0.0
pose_sensor/pose_noise_p_ic: 0.0
pose_sensor/pose_delay: 0.02
pose_sensor/pose_noise_meas_p: 0.005
pose_sensor/pose_noise_meas_q: 0.02
pose_sensor/pose_initial_scale: 5.60

# q_ic is the quaternion representing the rotation of the camera in IMU frame. Unit quaternion here as we rotate the coordinate frames in SVO parameters.
pose_sensor/init/q_ic/w: 1.0
pose_sensor/init/q_ic/x: 0.0
pose_sensor/init/q_ic/y: 0.0
pose_sensor/init/q_ic/z: 0.0

# p_ic is the translation between the IMU and the camera in meters.
pose_sensor/init/p_ic/x: 0.1 
pose_sensor/init/p_ic/y: 0.0
pose_sensor/init/p_ic/z: 0.065

pose_sensor/pose_absolute_measurements: true
pose_sensor/pose_use_fixed_covariance: true
pose_sensor/pose_measurement_world_sensor: false # we do not publish the world in camera frame as set in SVO parameters.

pose_sensor/pose_fixed_scale: false
pose_sensor/pose_fixed_p_ic: true
pose_sensor/pose_fixed_q_ic: true
pose_sensor/pose_fixed_p_wv: false
pose_sensor/pose_fixed_q_wv: false

I have a downward-facing camera with image top the same as pixhawk forward direction. So I think keep pose_sensor/init/q_ic/w:1 x:0 x:0 y:0 z:0 should be fine. But that can not make ATTITUDE.roll/pitch/yaw and VISION_POSITION_ESTIMATE.roll/pitch/yaw the same.
Could anyone help? Thanks!

2

Hi @rawk-v,

Sorry can’t help with your question. But I was wondering are you using mavros to feed in the vision_position_estimate (via mavros/vision_pose/pose)? and if so, do you need to adjust any parameter in LPE other than the vision_on and the corresponding weight?

Thank you

Best
Francis

3

@tuofeichen Yes, I use mavros to send vision data to FCU, but the topic is /mavros/vision_pose/pose_cov.
And I disable gps too in LPE estimator. Also you need to set ATT_EXT_HDG_M=1 if your want to fuse vision heading.

4

Hi @rawk-v,
Thanks, I figured what I did wrong by looking into mavros. It seems like we also need to assign the timestamp otherwise it won’t transmit the message with same timestamp. But thanks anyway!

5

Hello,

How did you manage to figure out a solution for your problem? What do you mean by assigning timestamp ?

6

Hello,

I know this question is a bit outdated, but would you happen to remember how to setup the MSF? I’ve been trying to look for a tutorial online but I haven’t had much luck. I want to try to use MSF with SVO for making a position hold controller.

Thank you!