parameters.md

Parameter reference

All parameters live in include/mod/parameters.hpp as plain structs with nlohmann JSON (de)serialization. from_json starts from the defaults and overrides only the keys present, so partial objects are accepted everywhere (batch JSON, --config, your own code). Enum values are the strings shown. Defaults equal the settings of the published experiments unless noted.

VehicleParameters — scope VehicleParameters

Key	Default	Meaning
shape	rectangle	circle or rectangle
radius	0.4	circle only [m]
length	0.7	rectangle only, along the heading [m] (the published robot spans x ∈ [-0.2, 0.5])
width	0.4	rectangle only [m]
state_space	dubins	dubins (forward only) or reeds_shepp (reverse allowed)
turning_radius	1.0	minimum turning radius [m]

The footprint test uses the circumscribed circle: radius, or sqrt((length/2)² + (width/2)²) = 0.403 m for the default rectangle. There is no resolution field: the collision step and the cost step are inferred (see Derived).

OptObjParameters — scope OptObjParameters

Key	Default	Meaning
type	cliff	cliff, gmmt, dtc, intensity, path_length
w_d	1.0	weight of the steering distance
w_q	1.0	weight of the heading-change term
w_c	0.1	weight of the MoD term (published: cliff 0.1, gmmt 0.1, intensity 0.2, dtc 0.02)
cliff_map_file	""	CLiFF-map XML (cliff, dtc)
gmmt_map_file	""	GMMT-map XML (gmmt)
intensity_map_file	""	intensity XML: the cost for intensity, a multiplier for cliff and dtc, and the default map of the samplers
max_vehicle_speed	1.0	dtc: the speed paired with the motion direction in the Mahalanobis distance [m/s]
mahalanobis_threshold	10.0	dtc: cap on the Mahalanobis distance per Gaussian
use_mixing_factor	true	dtc: weight each Gaussian's distance by its mixing factor

SamplerParameters — scope SamplerParameters

Key	Default	Meaning
type	iid	iid, ellipse, intensity, dijkstra, hybrid
bias	0.05	probability of the informed branch (intensity: the 1 − q draw; dijkstra and hybrid: the Dijkstra path)
dijkstra_cell_size	0.5	grid cell of the Dijkstra sampler [m]
hybrid_intensity_bias	0.01	hybrid: probability of the intensity branch
intensity_map_file	""	the samplers' intensity map; empty means the objective's
log_samples	false	write every draw to samples.json

PlannerParameters — scope PlannerParameters

Key	Default	Meaning
type	rrt_star	rrt_star, ait_star, hybrid_astar
max_planning_time	64.0	budget [s]; Hybrid A* returns earlier when it has its solution
seed	0	OMPL RNG seed for the run (0 is treated as 1; batches set it to seed0 + index)
range	0.0	RRT* steer range [m]; 0 lets OMPL choose (20 % of the space extent)
goal_bias	0.05	RRT*
batch_size	100	AIT* samples per batch
informed_sampling	true	RRT*: use the objective's informed sampler

HybridAStarParameters — scope HybridAStarParameters

Key	Default	Meaning
cell_size_m	0.25	search cell for duplicate detection and the heuristic grid [m]
angle_bins	72	heading bins for duplicate detection (5°)
primitive_length_m	0.0	arc length of a primitive; 0 means cell_size_m × √2
analytic_ratio	3.5	shot every floor(h_kin / (ratio × primitive length)) expansions (at least every expansion)
analytic_max_length_m	5.0	shots longer than this are not attempted
max_expansions	2000000	cap on expanded nodes
allow_reverse	true	reverse primitives; only effective under reeds_shepp (forced off under dubins; the effective value is written back to config.json)
change_penalty	1000.0	added to the cost per direction flip [cost units]; there is never a reverse penalty

Scenario — scope Scenario

Key	Default	Meaning
name	""	used in the run folder name
map_yaml	""	occupancy map (map_server yaml); relative paths resolve against the batch JSON or, in the GUI, the maps folder
start	[0, 0, 0]	[x, y, yaw]; yaw in [-π, π)
goal	[0, 0, 0]	[x, y, yaw]

Derived — scope Derived (written by the playground, never read)

Key	Meaning
occupancy_pixel_m	pixel size of the occupancy map
mod_cell_m	cell size of the objective's MoD map (CLiFF resolution, intensity cell; GMMT has none and falls back to the intensity map's, then the pixel)
collision_step_m	= pixel size; OMPL's validity resolution is pixel / extent
mod_cost_step_m	= min(mod_cell_m, occupancy_pixel_m); the objective's cost step
circumscribed_radius_m	the footprint radius actually used

RunMeta — scope RunMeta (written by the logger)

mod_version, git_hash (git describe --always --dirty --tags at configure time), hostname, started_at (ISO 8601, UTC, milliseconds).

RunConfig

The aggregate of all scopes above; config.json is its serialization:

{
  "VehicleParameters": {...}, "Derived": {...}, "SamplerParameters": {...}, "OptObjParameters": {...},
  "PlannerParameters": {...}, "HybridAStarParameters": {...}, "Scenario": {...}, "RunMeta": {...}
}

String helpers: MoD::to_string(enum) and MoD::shapeFromString, stateSpaceFromString, samplerTypeFromString, objectiveTypeFromString, plannerTypeFromString (throw std::invalid_argument on unknown text).