Advanced collision-detection

Collision-detection is a two-steps process. First the BroadPhase detects pairs of colliders that are potentially in contact or intersecting. Second, the NarrowPhase processes all these pairs in order to compute contacts points and generate collision events. Based on these points, the constraints solver computes forces that may generate contact force events.

All the pairs detected by the broad-phase are stored into two graph structures:

  • The contact graph stores all the potential contact pairs (between two non-sensor colliders) as well as the contact points generated by the narrow-phase.
  • The intersection graph stores all the potential intersection pairs (between a sensor collider and another collider) as well as the result of the boolean intersection test executed by the narrow-phase.

These two graphs are part of the NarrowPhase structure and are automatically updated by the PhysicsPipeline or the CollisionPipeline. Each node of these graph contains a collide handle and there is one graph edge per pair detected by the broad-phase.

Collision and contact force events#

The narrow-phase can generate collision events between two colliders. Each collision event is given optional flags:

  • CollisionEventFlags::SENSOR is set if at least one of the colliders involved in the collision is a sensor.
  • CollisionEventFlags::REMOVED is set if a collision stopped because at least one of the colliders involved in the collision was removed from the physics scene.

In addition, after forces are computed by the constraints solver, contact force events may be generated between two colliders subject to non-zero contact forces. Generally, the user isn’t interested in contact force events unless the force magnitudes exceed some threshold. In order to skip low-force events, the engine will compute the sum of the magnitude of all the contacts between the two colliders and only trigger a contact force event if that magnitude is larger than the threshold set with ColliderBuilder::contact_force_event_threshold or Collider::set_contact_force_event_threshold (defaults to 0) for any of the two colliders with the ActiveEvents::CONTACT_FORCE_EVENTS flag enabled.


Collision events (resp. contact force events) are only generated between two colliders if at least one of them has the ActiveEvents::COLLISION_EVENTS flag (resp. ActiveEvents::CONTACT_FORCE_EVENTS flags) in its active events.

The contact graph#

The contact graph can be read in order to determine whether two specific non-sensor colliders are in contact, or to determine all the non-sensor colliders in contact with one particular non-sensor collider. Contact points and contact normals will also be provided when a contact exists.

The contact geometry (contact points, contact normal, penetration depth, etc.) can be read from the contact manifolds stored in a contact pair:

  1. Each contact pair may contain multiple contact manifolds. Each contact manifold represents a set of contacts sharing the same contact normal.
  2. Each contact manifold contains the list of geometric contacts detected by the narrow-phase.
  3. Each contact manifold also contains a list of contacts that were processed by the constraints solver for force calculation (aka. the solver contacts). These solver contacts are a subset of the contacts detected by the narrow-phase, expressed in a way that is more efficient for the constraints solver to process. These solver contacts can be modified or deleted by the user using contact modification.

All the geometric contact data are expressed in the local-space of the colliders. The solver contacts are expressed in world-space.


Because the solver contacts can be modified by the user and are expressed in world-space, they are transients by nature: they are recomputed at each frame from the geometric contacts. Because of their transient nature, the constraint solver will store the forces it computes inside of the geometric contacts TrackedContact::data::impulse field instead of the solver contacts themselves.

Keep in mind that the contact graph contains one graph edge per pair detected by the broad-phase. So the fact that a contact pair can be found in the graph doesn't mean that the corresponding colliders are actually in contact (they may just be very close to one another, without touching). It is necessary to check either:

  • theContactPair::has_any_active_contact if you need to know if there exist at least one solver contact between the colliders.
  • the length of ContactManifold:points for each manifold in ContactPair::manifolds to determine if the colliders are really geometrically touching (independently from contact-modification).

There will always be only up to one contact manifold between two colliders with convex primitive shapes. If one collider has a shape composed of several pieces (trimesh, polyline, heightfield, or compound shape) then there will be multiple contact manifolds, one for each piece that may result in an actual contact.

Finally, keep in mind that the contacts and contact manifolds field names frequently end with a digit 1 or 2. For example contact_pair.manifolds[0].local_n1 and contact_pair.manifolds[0].local_n2. Fields ending with the digit 1 relate to the collider identified by contact_pair.collider1. Fields ending with the digit 2 relate to the collider identified by contact_pair.collider2.

In other words local_n1 is the contact normal expressed in the local space of the collider collider_pair.collider1, it points towards the exterior of the shape of collider_pair.collider1. On the other hand, local_n2 is expressed in the local space of the collider collider_pair.collider2 and points towards the exterior of the shape of collider_pair.collider2.


The contact pair returned by narrow_phase.contact_pair(handle1, handle2) does not necessarily have contact_pair.collider1 == handle1 && contact_pair.collider2 == handle2. It could be swapped: contact_pair.collider1 == handle2 && contact_pair.collider2 == handle1.

So keep that in mind when reading the contact information because it's contact_pair.collider1 and contact_pair.collider2 that determine to what collider the digits 1 and 2 relate in the contacts and contact manifolds fields.

The intersection graph#

The intersection graph can be read in order to determine whether two specific colliders (assuming at least one of them is a sensor) are intersecting, or to determine all the colliders intersecting one particular collider (assuming at least one collider of each pair is a sensor). The intersection graph contains one graph edge for each pair of colliders such that:

  1. At least one of the collider is a sensor.
  2. And they are close enough so the broad-phase considers they have a chance to be intersecting.

Each such edge contains one boolean indicating if the colliders are actually intersecting or not:


Keep in mind that intersection tests are performed between two colliders only if at least one of the colliders is a sensor. If they are both non-sensor colliders then they will be involved in the contact graph instead of the intersection graph.

Physics hooks#

Physics hooks are user-defined callbacks used to change the behavior of the physics simulation. In particular, they can be used to filter contacts (in a more flexible way than collision groups and solver groups) and to modify contacts before they are processed by the constraints solver.

Contact and intersection filtering#

Sometimes, collision groups and solver groups are not flexible enough to achieve the desired behavior. In that case, the contact filtering hooks let you apply custom rules to filter contact pairs and intersection pairs:

  • For each potential contact pair (between two non-sensor colliders) detected by the broad-phase, if at least one of the colliders involved in the pair has the bit ActiveHooks::FILTER_CONTACT_PAIRS enabled in its active hooks, then PhysicsHooks::filter_contact_pair will be called. If this filter returns None then no contact computation will happen for this pair of colliders. If it returns Some then the narrow-phase will compute contact points.
  • For each potential intersection pair (between a sensor colliders and another collider) detected by the broad-phase, if at least one of the colliders involved in the pair has the bit ActiveHooks::FILTER_INTERSECTION_PAIRS enabled in its active hooks, then PhysicsHooks::filter_intersection_pair will be called. If this filter returns false then no intersection computation will happen for this pair of colliders. If it returns true then the narrow-phase will test whether or not they are intersecting.

When PhysicsHooks::filter_contact_pair returns Some(flags) it needs to provide a set of solver flags for this contact pair. These solver flags indicate what happen with the contacts of this contact pair afterwards:

  • If the returned Some(flags) contains the SolverFlags::COMPUTE_IMPULSES bit, then the constraints solver will compute forces for these contacts. If this bit is not included in the returned flags, then no contact force will be computed for this pair of colliders.

Right now there is no solver flags other than SolverFlags::COMPUTE_IMPULSES. Other flags may be added in the future.

Contact modification#

It is possible to modify contacts after they have been computed by the narrow-phase. Contact-modification can have multiple advanced usages, for example:

  • The simulation of conveyor belts by modifying the tangent_velocity of solver contacts.
  • The simulation of one-way-platforms by deleting some contacts depending on the contact normal.
  • The simulation of colliders with non-uniform friction or non-uniform restitution coefficients, i.e., friction or restitution coefficients that depend on the contact points location.

The PhysicsHooks::modify_solver_contacts methods is called on each contact manifold between two colliders where at least one of them has the ActiveHooks::MODIFY_SOLVER_CONTACTS flag enabled in its active hooks.


Contact modification can be used to remove some (or all) solver contacts from a contact manifold. However, it cannot be used to add new contacts manually. If this is something that could useful to you, please consider openning an issue to let us know about your use-case so we can see if this is worth adding.

Contact-modification lets you change most characteristics of a contact, including the contact normal, contact friction/restitution coefficients, contact penetration depth, and warmstart impulses. None of these modifications are persistent (they are overwritten during the next timestep). There is one exception though: you can modify a user_data associated to each ContactManifold. This user_data will persist throughout timesteps as long as the ContactManifold remains alive (i.e. as long as some contacts exist between the touching parts of the colliders shapes). This can be useful to apply modification rules that depend on previous states of the contact (like whether or not this contact manifold existed during previous timesteps).

Continuous Collision Detection#

Continuous Collision Detection (CCD) is used to make sure that fast-moving objects don't miss any contacts (a problem usually called tunneling). See the rigid-body CCD section for details.