Lidar sensor calibration validation using voxelized planar approximation of point cloud

1 . A system comprising:

one or more processors; and

a non-transitory memory storing processor-executable instructions that, when executed by the one or more processors, cause the system to perform operations comprising:

receiving a current sensor alignment identifying a first position and first orientation of a first lidar device associated with a vehicle relative to a second position and second orientation of a second lidar device;

receiving, as a point cloud, first lidar data from the first lidar device and second lidar data from the second lidar device associated with an environment through which the vehicle is traversing;

associating the point cloud with voxels, wherein a first voxel of the voxels is associated with a first portion of the environment;

determining a first planar surface fit to lidar points of a first subset of the first lidar data associated with the first voxel;

determining a second planar surface fit to lidar points of a second subset of the second lidar data associated with the first voxel;

determining, as a first error associated with the first voxel, a first distance between a first lidar point of the first subset and the second planar surface;

determining, as a second error associated with the first voxel, a second distance between a second lidar point of the second subset and the first planar surface;

determining, based at least in part on the first error and the second error, a sensor alignment error associated with the current sensor alignment; and

either:

controlling the vehicle based at least in part on the current sensor alignment responsive to determining that the sensor alignment error is less than a threshold error; or

modifying, as a modified sensor alignment, the current sensor alignment based at least in part on determining that the sensor alignment error meets or exceeds the threshold error; and

controlling the vehicle based at least in part on the modified sensor alignment.

2 . The system of claim 1 , wherein the operations further comprise determining a subset of voxels from among the voxels to include for validating the current sensor alignment, wherein the first voxel is determined to be included in the subset of voxels based at least in part on at least one of:

determining that the first planar surface and the second planar surface are associated with a same feature of the environment,

determining that a first planarity score determined for the first subset of the first lidar data indicates that the first subset is planar,

determining that a number of lidar points in the first subset of the first lidar data meets or exceeds a threshold number of lidar points, or

determining that the first subset of the first lidar data is associated with a determined semantic classification.

3 . The system of claim 2 , wherein:

determining the first planarity score comprises at least one of:

determining a residual associated with fitting the first planar surface to the first subset of the first lidar data,

determining a variance of position of lidar points in the first subset in a direction orthogonal to the first planar surface, or

determining, by eigenvalue decomposition, eigenvalues associated with the first subset; and

determining that the first subset is planar comprises:

determining that the residual is less than a threshold residual,

determining that the variance is less than a threshold variance, or

determining that the eigenvalues include two eigenvalues that are positive or meet or exceed a first threshold number and a third eigenvalue that is zero or less than a second threshold number.

4 . The system of claim 2 , wherein:

the first voxel is determined based at least in part on determining to subdivide a second voxel larger than and containing the first voxel; and

the operations further comprise:

determining a second planarity score associated with a third subset of the first lidar data associated with the second voxel; and

determining that the second planarity score indicates that the third subset is non-planar,

wherein determining to subdivide the second voxel is based at least in part on determining that the first second score indicates that the third subset is non-planar.

5 . The system of claim 1 , wherein the operations further comprise determining a subset of voxels from among the voxels to include in a validation subset for validating the current sensor alignment, wherein:

determining the subset of voxels comprises excluding a second voxel from the subset of voxels based at least in part on one or more of:

determining that a third planar surface determined from a third subset of the first lidar data associated with the second voxel and a fourth planar surface determined from a fourth subset of the second lidar data associated with the second voxel are associated with different features in the environment,

determining that a planarity score determined for the third subset of the first lidar data indicates that the third subset is nonplanar,

determining that a number of lidar points in the third subset of the first lidar data is less than a threshold number of lidar points, or

determining that the third subset of the first lidar data is associated with a semantic classification that is disallowed for validation.

6 . The system of claim 1 , wherein determining the sensor alignment error comprises determining at least one of a total median angle or median angular error based at least in part on median angles determined for a subset of the voxels including the first voxel.

7 . One or more non-transitory computer-readable media storing processor-executable instructions that, when executed by one or more processors, cause one or more processors to perform operations comprising:

receiving a current sensor alignment of a first lidar device relative to a second lidar device;

receiving first lidar data from the first lidar device and second lidar data from the second lidar device;

associating the first lidar data and the second lidar data with voxels, wherein a first subset of the first lidar data is associated with a first voxel of the voxels and a second subset of the second lidar data is associated with the first voxel;

determining, for a first voxel of the voxels, a first geometric approximation associated with the first subset of the first lidar data;

determining, for the first voxel, a second geometric approximation associated with the second subset of the second lidar data;

determining, as a first error associated with the first voxel, a first distance between a first lidar point of the first subset and the second geometric approximation;

determining, as a second error associated with the first voxel, a second distance between a second lidar point of the second subset and the first geometric approximation; and

based at least in part on a sensor alignment error that is based at least in part on the first error and the second error, either:

determining to output a validation of the current sensor alignment based at least in part on determining that the sensor alignment error is less than an error threshold, or

determining to output an indication that the current sensor alignment is invalid based at least in part on determining that the sensor alignment error meets or exceeds the error threshold.

8 . The one or more non-transitory computer-readable media of claim 7 , wherein the operations further comprise determining a subset of voxels from among the voxels to include for validating the current sensor alignment, wherein the first voxel is determined to be included in the subset of voxels based at least in part on at least one of:

determining that the first geometric approximation and the second geometric approximation are associated with a same feature of an environment through which a vehicle is traversing,

determining that a first planarity score determined for the first subset of the first lidar data indicates that the first subset is planar,

determining that a number of lidar points in the first subset of the first lidar data meets or exceeds a threshold number of lidar points, or

determining that the first subset of the first lidar data is associated with a determined semantic classification.

9 . The one or more non-transitory computer-readable media of claim 8 , wherein:

determining the first planarity score comprises at least one of:

determining a residual associated with fitting the first geometric approximation to the first subset of the first lidar data,

determining a variance of position of lidar points in the first subset in a direction orthogonal to the first geometric approximation, or

determining, by eigenvalue decomposition, eigenvalues associated with the first subset; and

determining that the first subset is planar comprises:

determining that the residual is less than a threshold residual,

determining that the variance is less than a threshold variance, or

determining that the eigenvalues include two eigenvalues that are positive or meet or exceed a first threshold number and a third eigenvalue that is zero or less than a second threshold number.

10 . The one or more non-transitory computer-readable media of claim 8 , wherein:

the first voxel is determined based at least in part on determining to subdivide a second voxel larger than and containing the first voxel; and

the operations further comprise:

determining a second planarity score associated with a third subset of the first lidar data associated with the second voxel; and

determining that the second planarity score indicates that the third subset is non-planar,

wherein determining to subdivide the second voxel is based at least in part on determining that the first second score indicates that the third subset is non-planar.

11 . The one or more non-transitory computer-readable media of claim 7 , wherein the operations further comprise determining a subset of voxels from among the voxels to include in a validation subset for validating the current sensor alignment, wherein:

determining the subset of voxels comprises excluding a second voxel from the subset of voxels based at least in part on one or more of:

determining that a third geometric approximation determined from a third subset of the first lidar data associated with the second voxel and a fourth geometric approximation determined from a fourth subset of the second lidar data associated with the second voxel are associated with different features in an environment through which a vehicle is traversing,

determining that a planarity score determined for the third subset of the first lidar data indicates that the third subset is nonplanar,

determining that a number of lidar points in the third subset of the first lidar data is less than a threshold number of lidar points, or

determining that the third subset of the first lidar data is associated with a semantic classification that is disallowed for validation.

12 . The one or more non-transitory computer-readable media of claim 7 , wherein determining the sensor alignment error comprises determining at least one of a total median angle or median angular error based at least in part on median angles determined for a subset of the voxels including the first voxel.

13 . The one or more non-transitory computer-readable media of claim 12 , wherein determining at least one of the total median angle or the median angular error is based at least in part:

determining a first angle based at least in part on the first distance and a first distance from the first lidar device to the first voxel;

determining a second angle based at least in part on the second distance and a second distance from the second lidar device to the first voxel;

determining a first median angle based at least in part on the first angle and the second angle; and

at least one of determining the total median angle based at least in part on the first median angle and other median angles associated with other voxels of the subset of the voxels or determining the median angular error based at least in part on a difference between the first median angle and the other median angles.

14 . The one or more non-transitory computer-readable media of claim 7 , wherein:

the operations comprise modifying the current sensor alignment as a modified sensor alignment based at least in part on determining the indication that the current sensor alignment is invalid; and

modifying the current sensor alignment comprises altering at least one of a first position of the first lidar device, a first orientation of the first lidar device, a second position of the second lidar device, or a second orientation of the second lidar device to reduce to the sensor alignment error.

15 . A method comprising:

receiving a current sensor alignment of a first lidar device relative to a second lidar device;

receiving first lidar data from the first lidar device and second lidar data from the second lidar device;

associating the first lidar data and the second lidar data with voxels, wherein a first subset of the first lidar data is associated with a first voxel of the voxels and a second subset of the second lidar data is associated with the first voxel;

determining, for a first voxel of the voxels, a first geometric approximation associated with the first subset of the first lidar data;

determining, for the first voxel, a second geometric approximation associated with the second subset of the second lidar data;

determining, as a first error associated with the first voxel, a first distance between a first lidar point of the first subset and the second geometric approximation;

determining, as a second error associated with the first voxel, a second distance between a second lidar point of the second subset and the first geometric approximation; and

based at least in part on a sensor alignment error that is based at least in part on the first error and the second error, either:

determining to output a validation of the current sensor alignment based at least in part on determining that the sensor alignment error is less than an error threshold, or

determining to output an indication that the current sensor alignment is invalid based at least in part on determining that the sensor alignment error meets or exceeds the error threshold.

16 . The method of claim 15 , wherein the method further comprises determining a subset of voxels from among the voxels to include for validating the current sensor alignment, wherein the first voxel is determined to be included in the subset of voxels based at least in part on at least one of:

determining that the first geometric approximation and the second geometric approximation are associated with a same feature of an environment through which a vehicle is traversing,

determining that a first planarity score determined for the first subset of the first lidar data indicates that the first subset is planar,

determining that a number of lidar points in the first subset of the first lidar data meets or exceeds a threshold number of lidar points, or

determining that the first subset of the first lidar data is associated with a determined semantic classification.

17 . The method of claim 16 , wherein:

determining the first planarity score comprises at least one of:

determining a residual associated with fitting the first geometric approximation to the first subset of the first lidar data,

determining a variance of position of lidar points in the first subset in a direction orthogonal to the first geometric approximation, or

determining, by eigenvalue decomposition, eigenvalues associated with the first subset; and

determining that the first subset is planar comprises:

determining that the residual is less than a threshold residual,

determining that the variance is less than a threshold variance, or

determining that the eigenvalues include two eigenvalues that are positive or meet or exceed a first threshold number and a third eigenvalue that is zero or less than a second threshold number.

18 . The method of claim 15 , wherein the method further comprises determining a subset of voxels from among the voxels to include in a validation subset for validating the current sensor alignment, wherein:

determining the subset of voxels comprises excluding a second voxel from the subset of voxels based at least in part on one or more of:

determining that a third geometric approximation determined from a third subset of the first lidar data associated with the second voxel and a fourth geometric approximation determined from a fourth subset of the second lidar data associated with the second voxel are associated with different features in an environment through which a vehicle is traversing,

determining that a planarity score determined for the third subset of the first lidar data indicates that the third subset is nonplanar,

determining that a number of lidar points in the third subset of the first lidar data is less than a threshold number of lidar points, or

determining that the third subset of the first lidar data is associated with a semantic classification that is disallowed for validation.

19 . The method of claim 15 , wherein determining the sensor alignment error comprises determining at least one of a total median angle or median angular error based at least in part on median angles determined for a subset of the voxels including the first voxel.

20 . The method of claim 19 , wherein determining at least one of the total median angle or the median angular error is based at least in part:

determining a first angle based at least in part on the first distance and a first distance from the first lidar device to the first voxel;

determining a second angle based at least in part on the second distance and a second distance from the second lidar device to the first voxel;

determining a first median angle based at least in part on the first angle and the second angle; and

at least one of determining the total median angle based at least in part on the first median angle and other median angles associated with other voxels of the subset of the voxels or determining the median angular error based at least in part on a difference between the first median angle and the other median angles.