Intrinsics and extrinsics

[ubicray]

Hey! Firstly, thanks for great paper and code! I used this tool for intrinsics and extrinsics: https://github.com/hku-mars/LiDAR_IMU_Init I need help on understanding how to implement those values for DLIO:

Initialization result: Rotation LiDAR to IMU (degree) = -0.084620 -0.649203 -1.151916 Translation LiDAR to IMU (meter) = -0.056023 -0.021736 0.026451 Time Lag IMU to LiDAR (second) = -0.010057 Bias of Gyroscope (rad/s) = -0.007985 -0.012948 0.020327 Bias of Accelerometer (meters/s^2) = 0.010086 0.009814 0.010098 Gravity in World Frame(meters/s^2) = -0.027753 0.030412 -9.809914

Homogeneous Transformation Matrix from LiDAR to IMU: 0.999734 0.020119 -0.011298 -0.056023 -0.020101 0.999797 0.001704 -0.021736 0.011330 -0.001477 0.999935 0.026451 0.000000 0.000000 0.000000 1.000000

Refinement result: Rotation LiDAR to IMU (degree) = -0.014112 -0.341810 -1.171677 Translation LiDAR to IMU (meter) = -0.042382 -0.019593 0.020312 Time Lag IMU to LiDAR (second) = -0.010057 Bias of Gyroscope (rad/s) = -0.008959 0.001204 0.013092 Bias of Accelerometer (meters/s^2) = 0.013954 0.009286 -0.020293 Gravity in World Frame(meters/s^2) = -0.029866 0.030184 -9.791937

Homogeneous Transformation Matrix from LiDAR to IMU: 0.999773 0.020448 -0.005959 -0.042382 -0.020446 0.999791 0.000368 -0.019593 0.005965 -0.000246 0.999982 0.020312 0.000000 0.000000 0.000000 1.000000

Btw, using livox mid-360

[greymfm]

Homogeneous Transformation Matrix is defined as:

r11 r12 r13   t1
r21 r22 r23   t2
r31 r32 r33   t3
0   0   0     1

with R rotation matrix and t transformation vector Based on your calibration values, I would try something like this:

    imu:
    calibration: false
    intrinsics:
      accel:
        bias: [  0.013954, 0.009286, -0.020293 ]
        sm:   [ 1.,  0.,  0.,
                0.,  1.,  0.,
                0.,  0.,  1. ]
      gyro:
        bias: [  -0.008959, 0.001204, 0.013092 ]


  extrinsics:
    baselink2imu:
      t: [ -0.042382, -0.019593,   0.020312 ]
      R: [ 1.,         0.020448,   0.,
           -0.020446,  1.,         0.,
           0.,         0.,         1. ]
    baselink2lidar:
      t: [ 0.,  0.,  0. ]
      R: [ 1.,  0.,  0.,
           0.,  1.,  0.,
           0.,  0.,  1. ]

Maybe you can give some feedback what parameters you actually did use for the Livox MID-360 (what did work well for you)? ;-)

[85256638]

Homogeneous Transformation Matrix is defined as:

r11 r12 r13   t1
r21 r22 r23   t2
r31 r32 r33   t3
0   0   0     1

with R rotation matrix and t transformation vector Based on your calibration values, I would try something like this:

    imu:
    calibration: false
    intrinsics:
      accel:
        bias: [  0.013954, 0.009286, -0.020293 ]
        sm:   [ 1.,  0.,  0.,
                0.,  1.,  0.,
                0.,  0.,  1. ]
      gyro:
        bias: [  -0.008959, 0.001204, 0.013092 ]


  extrinsics:
    baselink2imu:
      t: [ -0.042382, -0.019593,   0.020312 ]
      R: [ 1.,         0.020448,   0.,
           -0.020446,  1.,         0.,
           0.,         0.,         1. ]
    baselink2lidar:
      t: [ 0.,  0.,  0. ]
      R: [ 1.,  0.,  0.,
           0.,  1.,  0.,
           0.,  0.,  1. ]

Maybe you can give some feedback what parameters you actually did use for the Livox MID-360 (what did work well for you)? ;-)

since the OP didn't report back on the above setting. I would like to share my settings: I'm using Livox Avia, the intrinsics and extrinsics are from: https://github.com/hku-mars/LiDAR_IMU_Init as below:

Initialization result:
Rotation LiDAR to IMU (degree)     = -1.313296 -1.126729 -0.301456
Translation LiDAR to IMU (meter)   =  0.016343  0.055562 -0.074086
Time Lag IMU to LiDAR (second)     = -0.003865
Bias of Gyroscope  (rad/s)         =  0.003193 -0.000242  0.003296
Bias of Accelerometer (meters/s^2) = -0.009860 -0.010175 -0.009963
Gravity in World Frame(meters/s^2) = -0.020684  0.188709 -9.808163

Homogeneous Transformation Matrix from LiDAR to IMU: 
 0.999793  0.005710 -0.019536  0.016343
-0.005260  0.999721  0.023021  0.055562
 0.019662 -0.022913  0.999544 -0.074086
 0.000000  0.000000  0.000000  1.000000


Refinement result:
Rotation LiDAR to IMU (degree)     = -1.321007 -0.742159  0.078993
Translation LiDAR to IMU (meter)   =  0.049017  0.023597 -0.070867
Time Lag IMU to LiDAR (second)     = -0.003865
Bias of Gyroscope  (rad/s)         = 0.002492 0.001597 0.000897
Bias of Accelerometer (meters/s^2) =  0.004518 -0.006717 -0.014002
Gravity in World Frame(meters/s^2) = -0.011651  0.187335 -9.802430

Homogeneous Transformation Matrix from LiDAR to IMU: 
 0.999915 -0.001080 -0.012980  0.049017
 0.001378  0.999734  0.023034  0.023597
 0.012952 -0.023050  0.999650 -0.070867
 0.000000  0.000000  0.000000  1.000000

I applied the above intrinsics and extrinsics into the dlio.yaml:

dlio:

  version: 1.1.1

  adaptive: true

  pointcloud:
    deskew: true
    voxelize: true

  imu:
    calibration: false
    intrinsics:
      accel:
        bias: [ 0.004518, -0.006717, -0.014002 ]
        sm:   [ 1.,  0.,  0.,
                0.,  1.,  0.,
                0.,  0.,  1. ]
      gyro:
        bias: [ 0.002492, 0.001597, 0.000897 ]

  extrinsics:
    baselink2imu:
      t: [ 0.049017, 0.023597, -0.070867 ]
      R: [ 0.999915,  -0.001080,  -0.012980,
           0.001378,  0.999734,  0.023034,
           0.012952,  -0.023050,  0.999650 ]
    baselink2lidar:
      t: [ 0.,  0.,  0. ]
      R: [ 1.,  0.,  0.,
           0.,  1.,  0.,
           0.,  0.,  1. ]

the result is rather shocking:

I only moved less than 5 meters, but the

Distance Traveled:  291.49 meters.
Distance to Origin: 181.76 meters.

I used the "rosservice call /robot/dlio_map/save_pcd 0.01 /home/test/ws/src/direct_lidar_inertial_odometry/PCD" to save the PCD file and it looks very off.

[85256638]

@kennyjchen I would be delighted if you could give some advice :)

[kennyjchen]

@85256638 The extrinsics seem relatively correct (looking at this). My guess is that it's an issue with the FOV of the Aria. DLIO's keyframing scheme really relies on the 360 nature of mechanical LiDARs. It should work for Livox Mid-360's though (I've tested that).