Skip to content

Parameters

Every parameter is reachable three ways under one name: as a control in the viewer (cargo run --release), as an attribute of a vinery_lab.VineyardParams fragment, and as a field of the matching Isaac Lab VineyardCfg fragment: params.<fragment>.<field> in Python and VineyardCfg(<fragment>=<Fragment>Cfg(<field>=...)) in Isaac Lab set the same thing. The slider range is what the viewer offers; Python takes any value the generator can build. Lengths are in meters.

Scene

SceneParams in Python, SceneCfg in Isaac Lab.

Scene-wide parameters, owned by no element: one seed and one date for everything on the ground.

Parameter Type Default Slider range Description
seed int 0 0 to 64 The one seed the whole scene is generated from. Every layer salts it with a constant of its own, so nudging one layer's knobs never re-rolls another: a new seed is a different vineyard, not a different trunk on the same one.
season float 0.5 0 to 1 Where in the growing season the scene is: 0.0 at budbreak, 1.0 at harvest. Today only the weeds read it, for which species are up and whether a bolter has bolted; the canopy does not yet.

Terrain

TerrainParams in Python, TerrainCfg in Isaac Lab.

The ground surface the scene stands on: hills the field is laid over, with tillage bumps riding on them.

length runs along X, the direction rows take at orientation 0, and width along Y. The hills are noise anchored in world space at feature_size, so a larger field shows more hills rather than larger ones, and their grade is capped by max_inclination. The bumps are a separate band, roughness tall and roughness_size long at the coarsest, that does not count towards the grade: adding them never flattens a hill.

Parameter Type Default Slider range Description
length float 80.0 5 to 200 Extent along X, in meters.
width float 50.0 5 to 200 Extent along Y, in meters.
max_inclination float 20.0 0 to 45 Upper bound on the hills' slope, in degrees. The elevation amplitude is solved from this and feature_size, so the same value gives the same steepness whatever the field's extent or resolution. This is the grade a route has to climb; roughness rides on top of it and is not counted here, the way a clod does not make a field steep.
feature_size float 16.0 2 to 60 Distance from one hill to the next, in meters. The noise field is anchored in world space at this size, so changing the extent uncovers more or less of the same landscape rather than rescaling it.
roughness float 0.08 0 to 0.5 Height of the bumps riding on the hills, in meters — the clods, ruts and tillage texture a machine rides over rather than climbs. Zero leaves the ground as bare hills.
roughness_size float 4.0 0.5 to 20 Longest wavelength in the bump band, in meters. Shorter octaves are added below it, down to the finest the grid resolves, so this is the coarsest bump rather than the only one.
detail int 32 2 to 64 Grid samples per feature: how finely the mesh follows the noise. Also the collision height field's resolution, and how short a bump may get, since the roughness band stops at the shortest wave the grid carries.

Parcel

ParcelParams in Python, ParcelCfg in Isaac Lab.

How vineyard rows are laid out across the terrain.

Solves the positions the planting puts a vine and a post at. vine_spacing also sizes the cordons the vines build, and trellis_height is how tall the posts stand.

Parameter Type Default Slider range Description
orientation float 0.0 -90 to 90 Row direction, in degrees counter-clockwise from +X.
headland float 6.0 0 to 20 Inset from the terrain's edge left unplanted, in meters — the turning area machinery needs at each end of a row.
row_spacing float 2.4 0.8 to 6 Distance between neighbouring row centerlines, in meters.
vine_spacing float 1.2 0.5 to 3 Distance between neighbouring vines along a row, in meters.
post_spacing float 6.0 2 to 12 Target distance between posts along a row, in meters. Panel count is solved from this and the row's actual length, so real post spacing comes out close to this value rather than exactly equal to it.
min_row_length float 10.0 0 to 50 Rows shorter than this, after clipping to the headland-inset rectangle, are dropped rather than planted.
trellis_height float 1.8 0.5 to 3 Height of the trellis above the ground, in meters: how tall the posts stand, and what the top catch wires hang just under.

Planting

PlantingParams in Python, PlantingCfg in Isaac Lab.

What stands on the ground, and where.

Each plant is authored as its own prim, /Vineyard/Planting/Row_000/Vine_007, so a simulator can address one: attach a semantic label, bind a rigid body, randomize it. A name refers to a planting slot, so a vine skipped by miss_rate leaves a gap in the numbering rather than shifting every name after it.

A slot that comes out young by young_rate is planted as a replant in its first season, one green shoot out of the bare ground rather than a shrunken mature vine, and young_scale says how much of a full-grown shoot the youngest of them has put out.

Parameter Type Default Slider range Description
miss_rate float 0.03 0 to 0.3 Fraction of planting positions left empty. Real vineyards have gaps, and a perception model trained without them learns that they can't happen.
young_rate float 0.08 0 to 0.5 Fraction of vines that are recent replants rather than mature.
young_scale float 0.55 0.2 to 1 How small the youngest replant is, relative to a full-grown shoot.

Pole

PoleParams in Python, PoleCfg in Isaac Lab.

One trellis post: a plain grey cylinder.

How tall a post is comes from the parcel's trellis_height, since the posts are what hold the wires up there, and where the posts stand from its post_spacing, so neither is here.

There is no variations either. A post is a manufactured object, and the only variety a row of them shows is in how each was driven: a centimeter off line, a degree off plumb, a few centimeters deeper. That is applied per placement and needs no geometry of its own.

Parameter Type Default Slider range Description
radius float 0.04 0.01 to 0.1 Post radius, in meters. The default is the 8 cm round softwood post that is the commonest thing in a European vineyard; a steel profile post is nearer half as thick.
sides int 8 3 to 16 Vertices around the post. The silhouette, and the only detail knob a straight tube has.

Wire

WireParams in Python, WireCfg in Isaac Lab.

The wires strung from post to post, and what the vines are trained onto.

One fruiting wire on the post axis at the vines' trunk_height, the head height the cordons are tied along, and above it catch_wires levels of pairs, one wire either side of the post, that the season's shoots grow up between. The levels are spread evenly from the fruiting wire to just under the parcel's trellis_height, so the top pair is the wire a hedger cuts to.

Each wire spans one panel, post to post, so a run follows the ground the way the posts do. There is no variations: a trellis is strung from one reel of wire, and every span is the same mesh stretched to its own length.

Parameter Type Default Slider range Description
catch_wires int 2 0 to 4 Levels of catch wires above the fruiting wire. Each level is a pair, one wire either side of the post, and the shoots grow up between them; two pairs is the usual vertical-shoot-positioned trellis.
radius float 0.0015 0.001 to 0.004 Wire radius, in meters. The default is the 3 mm high-tensile steel a trellis is strung with.

Vine

VineParams in Python, VineCfg in Isaac Lab.

The permanent woody framework of a grapevine.

A trunk rising to the head, one or two cordons running along the fruiting wire from there, and the spurs pruned back onto them. arms is 1 for a unilateral vine or 2 for a bilateral one; how far each cordon reaches is solved from the parcel's vine_spacing and cordon_gap rather than set directly.

Shape only: where vines stand, and which ones are missing or young, is the planting's.

Parameter Type Default Slider range Description
trunk_height float 0.9 0.3 to 1.6 Height of the fruiting wire above the ground at the posts, in meters, held at the top wire on posts too short for it. Each vine's trunk reaches the wire over it, so on uneven ground the trunks differ about this. Not the trellis height, which is where the tops of the posts are.
trunk_radius float 0.035 0.01 to 0.08 Trunk radius at the base, in meters.
trunk_wobble float 0.02 0 to 0.08 How far the trunk's axis wanders off vertical, in meters.
arms int 2 1 to 2 Cordons per vine: 1 for a unilateral vine, 2 for a bilateral one.
cordon_gap float 0.15 0 to 0.6 Bare wire left between the cordon tips of neighbouring vines, in meters. Together with the parcel's vine spacing this is what sets how far a cordon reaches.
cordon_radius float 0.022 0.008 to 0.05 Cordon radius at the head, in meters.
spur_spacing float 0.12 0.05 to 0.4 Target distance between spurs along a cordon, in meters.
spur_length float 0.05 0 to 0.15 How far a spur stands off its cordon, in meters.
shoots_per_spur float 1.8 0 to 3 Shoots per spur, as a fractional count: the whole part is certain and the fraction is the odds of one more. A spur is pruned to two buds, so 1.8 — two shoots four times in five, one otherwise — is what a healthy spur-pruned vine looks like.
roughness float 0.14 0 to 0.4 Depth of the bark ridges, as a fraction of the local radius.
sides int 8 3 to 16 Vertices around each tube. The silhouette, visible on every instance.
detail int 20 4 to 60 Rings per meter along each tube. Barely visible at row distance, so this is the cheaper of the two detail knobs to turn down.
variations int 32 1 to 64 How many distinct vine meshes the scene may hold. A budget, not a count: the plants are clustered and this is how many representatives the clustering may keep. Lower it to trade variety for memory, raise it to spend memory on variety.

Shoot

ShootParams in Python, ShootCfg in Isaac Lab.

One season's green growth off a spur.

Shoots are shaped here and placed by the vine: how many a spur pushes is the vine's shoots_per_spur, since that is a fact about its pruning rather than about a shoot. length is bud to tip, and whoever places a shoot varies it a little.

A shoot also carries the canopy, so the two leaf knobs live here rather than on the leaf: how many leaves a shoot holds and how they hang is a fact about the shoot. Leaf size is set nowhere; it comes out of each leaf's age, as a scale on a prototype of fixed area.

stray is the share of shoots the trellis failed to hold. A stray shoot leans out of the canopy, longer than the shoots beside it, and is exported as a deformable curve a physics engine bends; flexible off exports it as an ordinary static mesh at the same rest shape instead.

dormant turns the whole canopy into the winter's bare canes: brown, leafless, a bud at every node, and every one exported as a curve so a pruning robot can cut it.

Parameter Type Default Slider range Description
length float 0.75 0.1 to 1.6 Bud to tip, in meters — how tall a shoot stands above the spur it grew from. Whoever places one varies this a little per shoot.
radius float 0.006 0.002 to 0.015 Radius at the bud, in meters.
lean float 0.06 0 to 0.25 How far the tip wanders off vertical, in meters.
stray float 0.0 0 to 0.2 The fraction of shoots the trellis failed to hold: missed by shoot positioning, so grown out into the alley, or by hedging, so grown on past the top wire. A stray shoot leans out of the canopy and is exported as a deformable curve rather than a mesh, which makes this the most expensive knob in the scene.
flexible bool True Whether a stray shoot is exported as a deformable curve a solver bends. Off, it is a single static mesh at the same rest shape: the same lean out of the canopy, nothing for a solver to pick up. For a backend with no rods, or to keep a canopy's look without paying for the bodies.
skinned bool False Whether a shoot that bends is drawn as one skin bending with it, a mesh bound to every segment of its rod, rather than as a tube on each segment, which kinks at every joint. A skin bends only where a consumer poses it from the rod's bodies, as misina_lab.isaaclab does under Newton; anywhere else it keeps the shape it was built in.
segment float 0.09 0.02 to 0.2 Length of one segment of a bending shoot's rod, in meters — a stray shoot's or a dormant cane's.
dormant bool False Winter: every shoot is last season's cane after leaf fall — lignified brown, leafless, with a bud at every node — and is exported as a deformable curve a pruner can cut, whether the trellis holds it or not. internode is then the bud spacing, 10–15 cm on a dormant cane, and flexible off draws the canes as static meshes instead.
internode float 0.07 0 to 0.25 Distance between leaf nodes up the shoot, in meters — how many leaves it carries, said the way a viticulturist would. Zero leaves the shoot bare.
leaf_droop float 0.35 0 to 1.2 How far a full-grown blade pitches below horizontal, in radians. The small blades at the tip stand nearly straight out.
sides int 6 3 to 12 Vertices around the tube.
detail int 40 8 to 90 Rings per meter along the tube. Higher than anywhere else in the scene and cheaper than it looks: a stem is a shared mesh, and the bend at the tip needs the density.
variations int 4 1 to 8 How many distinct stem meshes the scene may hold. A budget, not a count: the shoots are clustered and this is how many representatives the clustering may keep, once for the shoots the trellis holds and again for the stray ones, which are clustered apart.

Leaf

LeafParams in Python, LeafCfg in Isaac Lab.

One blade of the canopy.

A blade is built from its veins, five lobes with a toothed margin, cut fresh for every mesh the budget keeps; lobing sets how deeply. Every blade is built at area, a full-grown leaf's, and a leaf's size comes from the scale it is placed at. Where leaves hang, how big each one ends up and how far it has grown is the shoot's; a leaf still growing near a shoot's tip is cut narrower, open at its base, folded and lighter green than it will be.

Parameter Type Default Slider range Description
lobing float 0.5 0 to 1 How deeply a full-grown blade lobes: 0 is an entire, heart-shaped leaf, 1 is cut deep into five lobes. Each leaf draws a little more or less.
curl float 1.0 0 to 2 How far a grown blade bends out of the flat shape it was cut as: its veins arching down off the petiole's top, the blade standing up a little between them, and its margin ruffled, all scaled together. The middle of a spread, with every leaf drawing its own share of it, and some cupping instead, their lobes lifting. Zero leaves a grown blade flat but for its fold.
fold float 0.15 -0.5 to 0.8 How far a grown blade folds up along its midrib, in radians each side: 0 holds it flat across, more folds it into a V and less bows it over the midrib. Cultivars differ: most are flat, some fold.
asymmetry float 0.5 0 to 1 How lopsided a blade may be: 0 for halves that are mirror images but for small differences, 1 for one half much the larger and the midrib curving away from it toward the apex. Each leaf draws its own share, and leans the other way from the leaf below it, as on a vine.
blade_angle float 0.8 0 to 1.5 How far a grown blade bends down off its petiole where they meet, in radians. A leaf just unfolded holds its blade in line with the stalk.
area float 0.015 0.005 to 0.04 The area of a full-grown leaf's blade and petiole, in m². A Vitis vinifera leaf runs 100 to 200 cm², by cultivar and vigour.
variations int 40 1 to 100 How many distinct blade meshes the scene may hold. A budget, not a count: the leaves are clustered by growth stage, lobing, curl and lean, and this is how many representatives the clustering may keep.
detail int 120 8 to 400 How finely a blade is meshed, as the number of triangles its area is cut into. The margin follows it: coarser drops the teeth first, then rounds off the sinuses, and keeps the lobes.

Cover

CoverParams in Python, CoverCfg in Isaac Lab.

What grows in the alley between two rows.

kind is the regime, by name, and the species are the generator's to pick from it. The sward is built as half-metre tiles of blades, instanced down each alley: variations is the budget of distinct tile meshes and detail the blades per square metre baked into one.

Parameter Type Default Slider range Description
kind str "spontaneous" none, spontaneous, sown, cereal What the alley grows, by name: none for a bare or tilled alley, spontaneous for a sward that came up on its own, ragged and gappy, sown for a drilled grass sward, one height and dense, or cereal for a winter rye in drill lines along the alley.
alternate bool False Leave every second alley bare: the cover on half the alleys, the tractor's tyres on the other half. The commonest permanent arrangement in France.
width float 0.75 0.1 to 1 How much of the alley's width the cover spans, as a fraction, centred on the alley. Three quarters keeps a band clear of the vines either side.
height float 0.15 0.02 to 1.5 Standing height, in meters: a few centimetres just after mowing, half a metre when left to head, more for a cereal in spring.
cover float 0.7 0 to 1 Fraction of the ground inside the band the cover actually covers.
dryness float 0.0 0 to 1 0.0 green, 1.0 straw — a Mediterranean alley in August.
variations int 12 1 to 64 How many distinct tile meshes the scene may hold. A budget, not a count.
detail int 600 50 to 2000 Blades per square metre baked into a tile at full cover. The one knob that trades sward density for triangles.

Weeds

WeedParams in Python, WeedCfg in Isaac Lab.

The plants that come up where nothing was sown: in the under-vine strip, and as escapes in the alley.

strip is what is done to the under-vine strip, and it picks the species; the scene's season tilts the mix between the spring and summer flushes and decides whether a bolter has bolted. Every plant is one mesh, bar the long stems flexible exports as deformable curves a physics engine bends: variations is the budget of distinct meshes and detail the triangles a leaf is cut into.

Parameter Type Default Slider range Description
strip str "mown" herbicide, tilled, mown, untouched What is done to the under-vine strip, by name, which picks the species: herbicide leaves the annual grasses and tall bolters a spray does not kill, tilled the annuals that come back from seed, mown the rosettes and tufts that duck the blade, untouched everything.
strip_width float 0.3 0.1 to 0.6 How far the under-vine strip reaches either side of the trunks, in meters.
pressure float 6.0 0 to 25 Plants per square metre in the strip. Zero is a clean strip; the ceiling is one plant per slot, twenty-five.
alley_pressure float 0.5 0 to 5 Plants per square metre in the alley, between the strips — the escapes.
tall float 0.3 0 to 1 The share of plants that are tall — bolters and broadleaves — as against low tufts, mats and rosettes.
flexible bool False Whether a stem at least 30 cm long — a bolter's wand, a broadleaf's runners — is exported as a deformable curve a solver bends, with its leaves riding it. Each is a chain of rigid bodies, ten to the metre: the default parcel at the defaults bends about 2,500 stems, 12,000 bodies.
skinned bool False Whether a stem that bends is drawn as one skin bending with it, a mesh bound to every segment of its rod, rather than as a tube on each segment, which kinks at every joint. A skin bends only where a consumer poses it from the rod's bodies, as misina_lab.isaaclab does under Newton; anywhere else it keeps the shape it was built in.
variations int 24 1 to 64 How many distinct plant meshes the scene may hold. A budget, not a count.
detail int 16 4 to 64 Triangles a leaf is cut into.