MULLS SLAM Parameter List

Parameters	Description	Default value
motion_compensation_method	method for motion compensation of lidar (0: disabled, 1: uniform motion model (from point-wise timestamp), 2: from azimuth, 3: from azimuth (rotation-only), 4: imu-assisted)	0
scan_to_scan_module_on	conduct scan-to-scan registration before scan-to-map registration or just conduct scan-to-map registration	false
initial_scan2scan_frame_num	only conduct scan to scan registration for the first　${initial_scan2scan_frame_num} frames when the feature points in the local map is not enough	3
used_feature_type	used_feature_type (1: on, 0: off, order: ground, pillar, beam, facade, roof, vetrex). ground, facade and roof points are planar points (using point-to-plane distance metric), pillar and beam are linear points (using point-to-line distance metric), vertex points are spherical points (using point-to-point distance metric).	"111111"
adaptive_parameters_on	use self-adaptive parameters for different surroundings and road situation	false
semantic_assist_on	apply semantic mask to assist the geometric feature points extraction, only available when the semantic segementation module is on or the semantic mask is provided	false
cloud_down_res	voxel size (m) of the voxel downsampling for the point cloud. if it is set as 0.0, then the voxel downsampling would be disabled	0.0
cloud_pca_neigh_r	pca neighborhood searching radius (unit: m) for target point cloud	0.5
cloud_pca_neigh_k	use only the k nearest neighbor in the r-neighborhood to do PCA	20
gf_grid_size	grid size (unit: m) for ground segmentation	2.0
gf_in_grid_h_thre	height threshold (unit: m) above the lowest point in a grid for ground segmentation	0.3
gf_neigh_grid_h_thre	height threshold (unit: m) among neighbor grids for ground segmentation	2.2
ground_normal_method	method for estimating the ground points' normal vector ( 0: directly use (0,0,1), 1: estimate normal in fix radius neighborhood , 2: estimate normal in k nearest neighborhood, 3: use ransac to estimate plane coeffs in a grid)	0
sharpen_with_nms_on	use non-maximum supression to get the sharpen feature points from unsharpen points or not (use higher threshold)	true
linearity_thre	pca linearity minimum threshold for target point cloud	0.65
planarity_thre	pca planarity minimum threshold for target point cloud	0.65
curvature_thre	pca local curvature threshold	0.1
fixed_num_downsampling_on	enable/disable the fixed point number downsampling (processing time's standard deviation would br smaller)	true
ground_down_fixed_num	fixed number of the detected ground feature points (for source point cloud)	500
pillar_down_fixed_num	fixed number of the detected pillar feature points (for source point cloud)	150
facade_down_fixed_num	fixed number of the detected facade feature points (for source point cloud)	350
beam_down_fixed_num	fixed number of the detected beam feature points (for source point cloud)	100
roof_down_fixed_num	fixed number of the detected roof feature points (for source point cloud)	0
unground_down_fixed_num	fixed number of the unground points used for PCA calculation	15000
corr_weight_strategy	weighting strategy for correspondences (1: on, 0: off, order: x,y,z balanced weight, residual weight, distance weight, intensity weight)	"1101"
z_xy_balance_ratio	the weight ratio of the error along z and x,y direction (only available when balanced weight is enabled)	1.0
pt2pt_res_window	residual window size for the residual robust kernel function of point to point correspondence (only available when residual weight is enabled)	0.1
pt2pl_res_window	residual window size for the residual robust kernel function of point to plane correspondence (only available when residual weight is enabled)	0.1
pt2li_res_window	residual window size for the residual robust kernel function of point to line correspondence (only available when residual weight is enabled)	0.1
reg_intersection_filter_on	filter the points outside the intersection aera of two point cloud during registration	true
normal_shooting_on	using normal shooting instead of nearest neighbor searching when determing correspondences	false
corr_dis_thre_init	distance threshold (unit: m) between correspondence points at begining (the first iteration). Increase this value when the vehicle is moving fast	1.5
corr_dis_thre_min	minimum distance threshold (unit: m) between correspondence points at begining	0.5
dis_thre_update_rate	update rate (divided by this value at each iteration) for distance threshold between correspondence points	1.1
converge_tran	registration convergence threshold for translation (unit: m)	0.001
converge_rot_d	registration convergence threshold for rotation (unit: deg)	0.01
post_sigma_thre	the maximum threshold for the posterior standard deviation of the least square adjustment during the registration (unit:m). if the posterior standard deviation is larger than this value, the registration would be regarded as failed.	0.25
reg_max_iter_num_s2s	max iteration number for icp-based registration (scan to scan)	15
reg_max_iter_num_s2m	max iteration number for icp-based registration (scan to map)	15
reg_max_iter_num_m2m	max iteration number for icp-based registration (map to map)	20
local_map_max_pt_num	max feature point number allowed for the local map	20000
local_map_max_vertex_pt_num	max vertex key point number allowed for the local map	1000
local_map_radius	the radius (unit: m) of the local map (regarded as a sphere aera)	100.0
local_map_recalculation_frequency	recalculate the linear features (pillar, beam points) in the local map every ${local_map_recalculation_frequency} frame	99999
append_frame_radius	the radius (unit: m) of the frame that used to append into the local map	60.0
apply_map_based_dynamic_removal	use map based dynamic object removal or not	false
dynamic_removal_radius	the radius (unit: m) of the map based dynamic object removing (only available when the dynamic object removal is on)	30.0
initial_guess_mode	use which kind of initial guess(0: no initial guess, 1: uniform motion(translation only), 2: uniform motion(translation+rotation), 3:imu based)	0
loop_closure_detection_on	do loop closure detection and pose graph optimization or not	false
submap_accu_tran	accumulated translation (unit: m) for generating a new submap	40.0
submap_accu_rot	accumulated rotation (unit: deg) for generating a new submap	120.0
submap_accu_frame	accumulated frame number for generating a new submap	150
robust_kernel_on	turn on the robust kernel function in pgo	true
cooling_submap_num	waiting for several submaps (without loop closure detection) after applying a successful pgo	2
equal_weight_on	using equal weight (identidy matrix) for the information matrix in pose graph optimization	false
diagonal_information_matrix_on	use diagonal information matrix in pgo or not	false
max_iter_inter_submap	max iteration number for inter submap pgo	100
max_iter_inner_submap	max iteration number for inner submap pgo	100
pose_graph_optimization_method	use which library to do pgo (select from g2o, ceres and gtsam)	"ceres"