Ray Tracing Mapping¶

Using ray tracing for mapping allows moving objects to be cleared while building up a global map.

The ray tracer takes point clouds from a lidar and uses the transform for the sensor to register the point cloud in the correct frame. For each point in a new point cloud, a ray is traced from the sensor origin to the point. Any voxel already in the global map that the ray passes through is a candidate to be cleared from the map.

How to run¶

dimos run mid360-pointlio-ray-trace --lidar-ip=<mid360-ip>

This blueprint runs the Mid-360 driver, Point-LIO, and the ray tracer, and visualizes with Rerun.

Input/Output¶

Inputs¶

Stream	Description
`lidar`	Lidar scans in sensor frame.
`tf`	Transform tree to register lidar scans.

Outputs¶

Stream	Description
`global_map`	Full point cloud of the global map.
`local_map`	Cylindrical subsection of the global map with height of the 5th to 95th percentile z coordinate and radius of the 95th percentile xy distance from the sensor origin.

Most downstream consumers can be built to use the local_map, allowing for incremental work only on parts of the map that possibly have changed.

Voxel representation¶

Each voxel contains the following data:

Field	Description
`health`	Occupancy score. Each lidar scan with a point in this voxel increments health and each scan that traces through it decrements health. Only voxels with positive health are considered occupied.
`support`	Number of healthy neighbors.
`num_pts`	Count of actual lidar hits binned to this voxel.
`sum`	Sum of all lidar points landing in this voxel, relative to the center of the voxel.
`m2`	Sum of the outer products of the points. Used for point covariance and normal fitting.
`normal`	The latest normal vector fit from the points binned to this voxel and its neighbors.
`next_fit_pts`	The number of points required before the normal vector will be refit on this voxel.
`fine`	A bitmask representing which micro-voxels are occupied.

Configuration¶

Depending on your requirements, you may want to tune the behavior of the ray tracer. Responsiveness, surface quality, and false positives/negatives can all be adjusted through the configuration.

The following config values are available:

Map resolution¶

Field	Description
`voxel_size`	The size of voxels, in m.
`fine_divisor`	How many micro-voxels per edge in each voxel. Example: 3 results in 27 micro-voxels per voxel.

Clearing properties¶

Field	Description
`min_health`	Minimum health of a voxel before it is deleted. Voxels spawn one above this health.
`max_health`	Maximum health of a voxel before it stops accumulating health.
`shadow_depth`	How far to continue tracing rays past their points, in m. Helps clear objects following or moving toward the sensor.
`grace_depth`	Voxels within this distance from points, in m, are spared from clearing. Helps with voxel clipping on floors and surfaces.
`graze_cos`	Spare voxels when the absolute value of the ray dot normal is below this value. Higher values clear only direct hits, lower values also clear grazing hits.

Performance¶

Field	Description
`max_range`	Maximum distance each ray will be traced, in m. Points beyond it are ignored. Used to limit computation.
`ray_subsample`	Trace every Nth ray. Used to limit computation.