Specifies which constraint solver Bullet will use to resolve collisions and constraints. Both solvers parallelize the workload, but differ in the strategy they use to do so. Parallel Gauss-Seidel (Islands) will be faster in cases that involve many small “islands” of interacting objects (for example, a large number of small separate book stacks), whereas Parallel Gauss-Seidel (Graph Coloring) should perform better when such “islands” are few and large (such as a huge collapsing building).

Although results obtained with these solvers will generally not be identical, qualitative differences should be minor.

The number of constraint iterations used by the solver. The more iterations, the more accurate the constraint and collision handling.

Allows the constraint solver to terminate before performing the full number of Constraint Iterations if it is close enough to the solution. Larger values can increase performance at the cost of accuracy.

Specifies that the constraints should be randomly reordered before each of the Constraint Iterations. This may improve stability, but incurs a minor performance hit.

Specifies that the solver should ensure that changes to an island of interacting objects (including adding, removing, or repositioning objects) do not cause other islands to produce different simulation results, unless those changes cause the objects to interact. Otherwise, the solver only guarantees that resimulating with the exact same input to the solver will produce the same results. Enabling this option may incur a minor performance hit, and may change the simulation results slightly.

Increasing the CFM (constraint force mixing) parameter will make contact constraints softer, and may increase the stability of the simulation. Contact constraints may be violated by an amount proportional to this parameter times the force that is needed to enforce the constraint.

Specifies what proportion of the constraint error for contact constraints will be fixed during the next simulation step. If ERP (error reduction parameter) is set to 0, constrained objects will drift apart as the simulation proceeds. If ERP is set to 1, the solver will attempt to fix all constraint error during the next simulation step (however, this may result in instability in some situations). A value between 0.1 and 0.8 is recommended for most simulations.

Tries to make interpenetrating objects split without adding velocity (to keep objects from explosively flying apart).

Errors when solving positional and velocity constraints can introduce some extra energy to the system. Although this option removes most of the extra energy, it degenerates quality a little bit, in particular for stable stacking.

Split Impulse only applies when objects interpenetrate by more than this distance. This number should be negative (representing less than 0 distance between the objects).

Overrides the Error Reduction Parameter for contact constraints where the penetration distance is within the Penetration Threshold and Split Impulse is enabled.

Specifies the constraint network where constraints produced by sticky collisions should be emitted. If not specified, the first constraint network encountered by the solver will be used. If no constraint networks exist, sticky collisions will be turned off.

This constraint network must be a relationship (i.e. Attach Internal Constraints to Object is not turned on), and will receive any constraints not added to an internal constraint network by Use Internal Constraint Networks.

If turned on, the solver will add constraints into the internal constraint network (ConstraintGeometry geometry subdata) for sticky collisions between two objects from the same DOP object. If the DOP object does not have ConstraintGeometry subdata, the constraint will be added to the External Constraint Network. An internal constraint network (ConstraintGeometry subdata) can be added to an object using a Constraint Network DOP with Attach Internal Constraints to Object turned on.

The RBD Solver will break a full timestep into at least this number of substeps.

By increasing this, you can guarantee a minimum fineness to the substepping. This can be used if for some reason the automatic computations are too coarse.

The RBD Solver will not break the simulation down into more substeps than this.

It is a very good idea to always have a maximum to ensure frames will be finished regardless of their complexity. Lowering this ceiling can ensure a minimum computation time at the expense of accuracy.

The CFL Condition is a factor used for automatically determining what size substep a scene requires. The idea is that any substep should not allow any objects to interpenetrate by more than one voxel cell.

This condition is met when this parameter is at 1. A value of 10 would allow a substep to interpenetrate by as many as 10 voxel cells. This could allow objects to tunnel through each other rather than properly bounce.

The stack solver iterates over all objects looking for ballistic collisions. Because resolving one collision may create a collision elsewhere, this cannot be resolved in a single pass with a local solution.

The stack solver will thus repeat the collision resolution until either no collisions are found, or this pass count is reached.

Even if a collision is not fully resolved with these passes, it will still be cleaned up in the Contact Pass. The main difference is that it will become inelastic.

The stack solver iterates over all objects, looking for cases where resting contact requires an acceleration to be adjusted.

Multiply stacked objects are common, so this often has complicated interrelationships, so requires multiple passes to converge.

Resting objects have a higher stability requirement than bouncing objects. Thus, the object is not immediately brought to a standstill, but slowed over multiple iterations to allow the system to stabilize.

This is the number of steps to do this for every contact pass.

These passes are very similar to Contact Passes.

The main difference is that if a book were resting on a table, the table would be assigned infinite mass in this pass. This prevents the table from shifting into the ground, allowing the system to converge faster.

As a rule of thumb, set this to the expected maximum number of stacked objects. If you plane to have ten tables stacked on top of one another in a stable configuration, a value of 10 can help ensure that the stacking is fully resolved.

If your objects come to rest appropriately, but then seem to slowly start to sink through each other, increasing Shock Propagation can be the right answer.

These passes are a final attempt to prevent any interpenetration. Like Shock Propagation, it is attempted to process objects from bottom up.

If a book is resting on a table and is penetrating the table, the book will be moved to lie outside of the table. This will be performed even if the book is at rest on the table.

The penetration recover repeated until there are no more penetrations up to the maximum number provided by this parameter.

The SubContact Passes is used to slowly feather the objects apart. Rather than immediately moving the book outside of the table, it is done in over the given number of subcontact passes. This is done to attempt to stabilize the process when complicated overlaps occur.

Determines if changes in the point positions will be used in collision resolution. Note that this is different from the Inherit Velocity option of RBD State. This flag only governs if velocity attributes are used for collisions, not for setting up the initial velocities.

When this is set, the object is inspected for any per point velocity attribute. If present, it is assumed to be a local deformation vector and is used to improve collision response.

If no point velocities are present, the geometry is compared between the two frames to manually calculate the per-point velocity. Note that if your deformation is a function of $F you may not get expected results as that is a step function, use $T instead.

Determines if changes to the volumetric representation will be used in collision resolution.

When this is set the volumetric representation is compared between this frame and the previous frame. The difference is used to compute a velocity of the surface’s deformation. This allows deforming objects to interact plausibly.

discover tangential deformational velocities.

When objects are resting on top of one another, they still receive impacts due to the force of gravity. This option prevents these from being added to the glue impulse, making it easier to prevent things from falling apart under their own gravity.

During the RBD solving process, numerous impacts are calculated between the RBD objects. These are normally not recorded in order to save time.

If this is set, however, all such impacts will be recorded by attaching an Impacts/RBDImpacts data to the objects that collide.

In simulations with a large number of objects, it is helpful to use various space partitioning schemes to reduce the work in finding collisions. This option selects one of these schemes.

None means that no attempt at spatial subdivision will occur.

Sphere means the objects will be treated as spheres and trivial intersection detection will be done with these spheres. This is fast, but with long skinny objects could cause false positives.

OBB means Oriented Bounding Boxes. While this provides a tight bound on long skinny objects, building the spatial partitioning tree is slow and will often exceed the benefits.

Controls whether and how Houdini groups similar points together when it calculates point collisions.

If you set this parameter to a value other than “none”, Houdini will treat similar points (that is, points within the distance specified in the Contact grouping tolerance below) as a single point for the purpose of calculating collisions.

This is useful when you have an object such as a cube, where the geometry points (the corners of the cube) are spaced far apart. One corner might impact a ground plane first, then the cube bounces and rotates so the opposite corner hits, which bounces and rotates, causing jitter when the cube hits.

If you set the Contact Grouping Method to “Average”, Houdini will calculate the hit based on an average point between the corners, giving a more stable result with less jitter.

This is similar to the effect of turning on Edge representation in the Surface tab of an RBD Object node. If you have sparse geometry with sharp edges, such as a cube, you may want to turn on both these options.

To see the effect of contact groupings, create a simulation where you drop a cube onto a ground plane. Attach an RBD Visualization DOP to see the resulting impacts.

The distance within which points are grouped together when Contact grouping method is not “none”.

The minimum volume for any piece geometry created by this solver. This can avoid creating geometry that is too small for the RBD Solver to handle in a stable manner.

The tolerance to use when fusing clustered pieces together.

Creates an attribute called creationtime on all newly-created interior primitives that stores their creation time. This can be useful for effects such as emitting particles for debris for a short amount of time after the initial fracture of a primitive.

Usually fracturing is driven only by objects that solve before this object in the frame. If two-way coupling between different solvers is done using feedbacks, this option allows those feedback effects to also trigger fracturing.

Impacts generated by objects merely resting on each other will not be considered for fracturing.

This value is used to initialize the pseudo-random sequence used to generate fracture points during the simulation. It is useful for generating different simulations from the same network.

The fracturing only works with polygonal geometry. This option auto-converts the geometry into polygons at the given LOD. This conversion only happens when the fracturing occurs.

Each data option parameter has an associated menu which specifies how that parameter operates.

For any parameters with their Operation menu set to Use Default, this parameter controls what operation is used.

This parameter has the same menu options and meanings as the Parameter Operations menus, but without the Use Default choice.

All objects connected to the first input of this node become mutual affectors.

This is equivalent to using an Affector DOP to create an affector relationship between * and * before connecting it to this node. This option makes it convenient to have all objects feeding into a solver node affect each other.

When an object connector is attached to the first input of this node, this parameter can be used to choose a subset of those objects to be affected by this node.

Indicates the name that should be used to attach the data to an object or other piece of data. If the Data Name contains a “/” (or several), that indicates traversing inside subdata.

For example, if the Fan Force DOP has the default Data Name “Forces/Fan”. This attaches the data with the name “Fan” to an existing piece of data named “Forces”. If no data named “Forces” exists, a simple piece of container data is created to hold the “Fan” subdata.

Different pieces of data have different requirements on what names should be used for them. Except in very rare situations, the default value should be used. Some exceptions are described with particular pieces of data or with solvers that make use of some particular type of data.

Turning on this parameter modifies the Data Name parameter value to ensure that the data created by this node is attached with a unique name so it will not overwrite any existing data.

With this parameter turned off, attaching two pieces of data with the same name will cause the second one to replace the first. There are situations where each type of behavior is desirable.

If an object needs to have several Fan Forces blowing on it, it is much easier to use the Unique Data Name feature to ensure that each fan does not overwrite a previous fan rather than trying to change the Data Name of each fan individually to avoid conflicts.

On the other hand, if an object is known to have RBD State data already attached to it, leaving this option turned off will allow some new RBD State data to overwrite the existing data.

The default behavior for solvers is to attach the exact same solver to all of the objects specified in the group. This allows the objects to be processed in a single pass by the solver, since the parameters are identical for each object.

However, some objects operate more logically on a single object at a time. In these cases, one may want to use $OBJID expressions to vary the solver parameters across the objects. Setting this toggle will create a separate solver per object, allowing $OBJID to vary as expected.

Setting this is also required if stamping the parameters with a Copy Data DOP.