Collider components define the shape of an object for the purposes of physical collisions. A collider, which is invisible, need not be the exact same shape as the object’s mesh and in fact, a rough approximation is often more efficient and indistinguishable in gameplay. The simplest (and least processor-intensive) colliders and the so-called primitive collider types. In 3D, these are the Box Collider, Sphere Collider and Capsule Collider. In 2D, you can use the Box Collider 2D and Circle Collider 2D. Any number of these can be added to a single object to create compound colliders. With careful positioning and sizing, compound colliders can often approximate the shape of an object quite well while keeping a low processor overhead. Further flexibility can be gained by having additional colliders on child objects (eg, boxes can be rotated relative to the local axes of the parent object). However, you should be sure that there is only one Rigidbody and this should be placed on the root object in the hierarchy.
There are some cases, however, where even compound colliders are not accurate enough. In 3D, you can use Mesh Colliders to match the shape of the object’s mesh exactly. In 2D, the Polygon Collider 2D will generally not match the shape of the sprite graphic perfectly but you can refine the shape to any level of detail you like. These colliders are much more processor-intensive than primitive types, however, so use them sparingly to maintain good performance. Also, a mesh collider will normally be unable to collide with another mesh collider (ie, nothing will happen when they make contact). You can get around this in some cases by marking the mesh collider as Convex in the inspector. This will generate the collider shape as a “convex hull” which is like the original mesh but with any undercuts filled in. The benefit of this is that a convex mesh collider can collide with other mesh colliders so you may be able to use this feature when you have a moving character with a suitable shape. However, a good general rule is to use mesh colliders for scene geometry and approximate the shape of moving objects using compound primitive colliders.
Colliders can be added to an object without a Rigidbody component to create floors, walls and other motionless elements of a scene. These are referred to as static colliders. In general, you should not reposition static colliders by changing the Transform position since this will impact heavily on the performance of the physics engine. Colliders on an object that does have a Rigidbody are known as dynamic colliders. Static colliders can interact with dynamic colliders but since they don’t have a Rigidbody, they will not move in response to collisions.
The reference pages for the various collider types linked above have further information about their properties and uses.
When colliders interact, their surfaces need to simulate the properties of the material they are supposed to represent. For example, a sheet of ice will be slippery while a rubber ball will offer a lot of friction and be very bouncy. Although the shape of colliders is not deformed during collisions, their friction and bounce can be configured using Physics Materials. Getting the parameters just right can involve a bit of trial and error but an ice material, for example will have zero (or very low) friction and a rubber material with have high friction and near-perfect bounciness. See the reference pages for Physic Material and Physics Material 2D for further details on the available parameters. Note that for historical reasons, the 3D asset is actually called Physic Material (without the S) but the 2D equivalent is called Physics Material 2D (with the S).
The scripting system can detect when collisions occur and initiate actions using the OnCollisionEnter function. However, you can also use the physics engine simply to detect when one collider enters the space of another without creating a collision. A collider configured as a Trigger (using the Is Trigger property) does not behave as a solid object and will simply allow other colliders to pass through. When a collider enters its space, a trigger will call the OnTriggerEnter function on the trigger object’s scripts.
When collisions occur, the physics engine calls functions with specific names on any scripts attached to the objects involved. You can place any code you like in these functions to respond to the collision event. For example, you might play a crash sound effect when a car bumps into an obstacle.
On the first physics update where the collision is detected, the OnCollisionEnter function is called. During updates where contact is maintained, OnCollisionStay is called and finally, OnCollisionExit indicates that contact has been broken. Trigger colliders call the analogous OnTriggerEnter, OnTriggerStay and OnTriggerExit functions. Note that for 2D physics, there are equivalent functions with 2D appended to the name, eg, OnCollisionEnter2D. Full details of these functions and code samples can be found on the Script Reference page for the MonoBehaviour class.
With normal, non-trigger collisions, there is an additional detail that at least one of the objects involved must have a non-kinematic Rigidbody (ie, >Is Kinematic must be switched off). If both objects are kinematic Rigidbodies then OnCollisionEnter, etc, will not be called. With trigger collisions, this restriction doesn’t apply and so both kinematic and non-kinematic Rigidbodies will prompt a call to OnTriggerEnter when they enter a trigger collider.
You can attach one rigidbody object to another or to a fixed point in space using a Joint component. Generally, you want a joint to allow at least some freedom of motion and so Unity provides different Joint components that enforce different restrictions. For example, a Hinge Joint allows rotation around a specific point and axis while a Spring Joint keeps the objects apart but lets the distance between them stretch slightly. As usual, the 2D components have 2D at the end of the name, eg, Hinge Joint 2D.
Joints also have other options that can enabled for specific effects. For example, you can set a joint to break when the force applied to it exceeds a certain threshold. Some joints also allow a drive force to occur between the connected objects to set them in motion automatically.
See the reference pages for the Joint classes to read further about their properties.
The character in a first- or third-person game will often need some collision-based physics so that he doesn’t fall through the floor or walk through walls. Usually, though, the character’s acceleration and movement will not be physically realistic, so he may be able to accelerate, brake and change direction almost instantly without being affected by momentum.
In 3D physics, this type of behaviour can be created using a Character Controller. This component gives the character a simple, capsule-shaped collider that is always upright. The controller has its own special functions to set the object’s speed and direction but unlike true colliders, a rigidbody is not needed and the momentum effects are not realistic.
A character controller cannot walk through static colliders in a scene, and so will follow floors and be obstructed by walls. It can push rigidbody objects aside while moving but will not be accelerated by incoming collisions. This means that you can use the standard 3D colliders to create a scene around which the controller will walk but you are not limited by realistic physical behaviour on the character itself.
You can find out more about character controllers on the reference page.