Splintering a Wooden Fence With Rigid Body Simulations | Houdini Tutorial

Break a localized section of a wooden fence by launching a rigid-body boulder through it. The central modeling technique is to compress the fence section before Voronoi fracturing and restore its scale afterward, stretching the resulting pieces into splinters. A second set of painted-density scatter points adds smaller fragments around the impact area.

The lesson starts with an existing boulder, fence and ground, then uses Houdini’s historical shelf-driven DOP workflow. Their asset construction and the exact Houdini version are outside this recording. The final simulation is an artistic example with acknowledged intersections and additional off-camera tweaks, not a collision-clean production setup.

Establish the boulder trajectory before fracturing

At 01:01, make the ground a Static Object. The shelf builds an AutoDopNetwork containing the imported ground, static solver, gravity and output. Make the boulder an RBD Object so it joins the dynamic side of that network.

At 02:01, give the boulder initial velocity toward the fence and a downward component so it bounces before impact. The narration tries a forward value of 50 and corrects the downward component’s sign. Its spoken axis descriptions are not consistent enough to reproduce an exact velocity vector from captions alone; verify direction against the actual scene axes and playback.

At this stage the boulder passes through the intended fence region because the fence dynamics are not set up yet. Use the trajectory to identify the impact area and approximate contact time, not as proof that the fence already responds physically.

Isolate the region that will break

At 04:02, inspect the fence’s geometry construction. A former Merge has been replaced with a Boolean so the fracture operates on joined geometry. Add a coarse Voronoi Fracture driven by a small scatter count, initially around five.

Use Exploded View to inspect the resulting regions and change the scatter seed until one covers the desired damage area. Remove the exploded display from the working result once it has served that inspection purpose.

At 06:01, identify the chosen piece with Blast. The example finds the desired region at the piece numbered four, then duplicates the Blast setup: one branch retains the surrounding fence, while Delete Non Selected on the other retains the breakable region. This identifier is specific to the chosen fracture and seed. Check the pieces again if either changes.

Keep these complementary branches available. One supplies the detailed fracture; the other provides the intact-looking surroundings, whose collision participation is tested separately later.

Stretch the fracture pattern into splinters

At 08:00, an ordinary fracture of the selected region makes chunky pieces that resemble concrete more than wood. Reduce the initial excessive point count and try roughly 300 points, then change the space in which the fracture is calculated.

At 09:00, scale the section to 0.1 along Y before generating the fracture inputs. After fracturing, scale the result by ten along the same axis. These reciprocal transforms return the fence section to its original height while elongating the internal fracture cells into vertical splinters.

Inspect with Exploded View again. The intended change is the pieces’ proportions, not an accidental change in the assembled fence’s overall size. The compressed geometry and the points used to cut it must occupy the same working space.

At 10:01, add IsoOffset to supply an interior volume for scattering. The demonstrated result has more internal pieces than the previous setup. Increase its usable definition as needed in the shown scene, then judge whether the fragments resemble the desired wood breakup.

Add smaller fragments around the impact

At 11:00, color the isolated fence region black and paint white around the intended boulder contact area. Inspect the underside as well so the painted region covers the relevant surfaces.

Scatter a second set of points using the painted color as its density attribute. Around 13:01, the example uses roughly 500 points for this localized contribution and merges them with the broader fracture-point set.

This new branch also needs the same compression as the rest of the fracture setup. The instructor adds a matching transform before merging the point sets. Without that relationship, the painted detail points would describe a different spatial arrangement from the compressed fence being cut.

The combined pattern produces smaller pieces in the dense impact region and larger splinters elsewhere. The merge reports mismatched attributes, but the instructor accepts the displayed calculation. That observation is specific to the demonstrated geometry; it is not a general instruction to ignore attribute warnings in a different setup.

Separate visible surroundings from their collision role

At 14:03, use the RBD Packed shelf setup on the fractured fence. Restart from frame one and play. The pieces respond, but they also fall under gravity before the boulder arrives. The instructor identifies contact near frame 11 and tries an activation approach around frame 10, then abandons it. This is not the final working activation recipe.

Merging the surrounding fence into the dynamic geometry creates a different problem: too much of the fence participates in the motion and obstructs the boulder. Density and bounce adjustments are explored, but the whole surrounding region should not simply fly away with the impact pieces.

At 19:00, expose the surrounding branch with an output null and import it into a separate geometry object through Object Merge. It initially appears missing. The investigation finds it displaced elsewhere in the scene; parenting the objects together in this setup restores the intended alignment around 22:00. The source’s proposed restart is not the established fix for that problem.

Test this surrounding fence as a Static Object. It prevents some overlap, but the instructor prefers the breakup when those surrounding pieces do not collide with the dynamic fragments. The final chosen approach removes that static collision setup while retaining the surrounding geometry visually.

This decision has a visible cost: dynamic pieces can pass into that non-colliding fence. Keep the distinction explicit when judging the result. The unbroken-looking surroundings are not evidence that all displayed surfaces are participating in the solver.

Use glue and contact settings to tune the breakup

At 24:02, apply Glue Adjacent to the fractured fence. The generated constraints initially hold the pieces too strongly. Lower their strength and replay from the start, checking two different conditions: the fence should remain together before contact, then break when struck.

The source tries strengths including 0.1 and 0.01. These are example values for this particular geometry, mass and trajectory, not universal wood settings. Gluing pieces to one another also should not be described as a separate world-anchoring system; no such anchoring construction is demonstrated here.

At 28:01, adjust fence bounce, friction and a control described as rotational stiffness. Some settings make shards move like arrows toward the camera. Excessive friction later causes an unwanted grabbing response, so it is reduced. Ground bounce and friction trials are also revisited rather than all being kept.

Shorten the sequence to about 96 frames around 31:02. Then test the surrounding static collider again: it prevents one overlap but causes awkward backward motion. Making the glue too weak introduces premature collapse. The instructor removes the static collision setup again and settles on a later glue-strength adjustment from 0.05 to about 0.08 near 34:03.

The final preview shows the fence breaking into shards and splinters, with some further tweaks made off camera. Verify the trajectory, pre-impact stability, concentrated small fragments, elongated pieces and ground response separately. The lesson demonstrates how fracture geometry and simulation tuning work together, while leaving collision cleanup and an exact final parameter inventory unresolved.