Unity Physics Integration: Collider Tuning & Layer Matrix Setup

Incorrectly configured colliders are one of the most common and subtle problems in Unity projects. We, as engineers with 10+ years of production experience, see it constantly: the character falls through the floor at high speed, bullets vanish without hitting the target, physics objects jitter on a static surface. The causes are almost always fundamentals: wrong collider type, missing Continuous collision detection, incorrect layer settings. Our team guarantees to eliminate these issues turnkey — contact us, we'll evaluate your project.

Why colliders are set up incorrectly?

Unity offers six primitive colliders: BoxCollider, SphereCollider, CapsuleCollider, MeshCollider, WheelCollider, TerrainCollider. Plus 2D equivalents. MeshCollider is the main source of problems in inexperienced hands. Convex MeshCollider works correctly with Rigidbody but is limited to 255 polygons. Non-convex cannot be applied to dynamic objects at all — Unity will give a warning but won't block it, leading to undefined behavior. For characters and projectiles we always use primitives; MeshCollider only for static level geometry.

CapsuleCollider is optimal for characters (vertical) and bullets (horizontal). Two parameters: radius and height. Typical mistake: the character's capsule is sized to the visual mesh instead of gameplay needs — too wide makes the character 'fatter' than it looks, so it can't fit through narrow passages. SphereCollider is the cheapest computationally. For projectiles, grenades, and small items, prefer CapsuleCollider if the shape allows.

How to set up Layer Matrix?

Physics Layer Collision Matrix (Edit → Project Settings → Physics → Layer Collision Matrix) determines which layers interact. An incorrect matrix leads to: projectiles hitting their own team; trigger zones reacting to the environment instead of only the player; enemies pushing each other when grouped. A proper layer structure for a typical game: Default, Player, Enemy, Projectile, Environment, Trigger, Debris. Projectile interacts with Player, Enemy, Environment — but not with Trigger, Debris, other Projectile. Trigger has no physical interaction — only OnTriggerEnter. When using Physics.Raycast or Physics.OverlapSphere, always pass LayerMask explicitly — without a mask, the cast checks all layers, including invisible UI colliders and trigger zones, returning unexpected hits.

Collision Detection Mode: defeating the tunnelling effect

By default, Rigidbody uses Discrete collision detection: the object's position is checked at the beginning and end of each FixedUpdate. If the object moves faster than objectSize / fixedDeltaTime units per second, it can 'jump' through thin geometry — this is the tunnelling effect. A bullet of diameter 0.1 units at speed 50 m/s travels 50 * 0.02 = 1 unit per one FixedUpdate. A wall of thickness 0.5 units will be missed. Solutions:

• Rigidbody.collisionDetectionMode = CollisionDetectionMode.Continuous — for dynamic objects that may tunnel through static objects
• CollisionDetectionMode.ContinuousDynamic — for objects that may tunnel through other dynamic objects
• For extremely high-speed bullets, use Physics.Raycast or Physics.SphereCast instead of a physical Rigidbody: more reliable and performant

Continuous mode is more CPU-intensive. Apply only to fast projectiles and characters — not to all objects indiscriminately.

PhysicMaterial and friction tuning

PhysicMaterial defines dynamicFriction, staticFriction, and bounciness. The combination of two materials from contacting objects follows frictionCombine and bounceCombine rules (Average, Minimum, Multiply, Maximum). For a character on Rigidbody: PhysicMaterial with dynamicFriction = 0, staticFriction = 0, frictionCombine = Minimum. Without this, the character slips along walls, gets stuck on edges, and unexpectedly slows on slopes. For bouncing objects: bounciness = 0.6, bounceCombine = Maximum. With bounceCombine = Average, a ball thrown onto a surface with bounciness = 0 won't bounce at all — even if the ball itself has a high value.

Compound colliders and optimization

We describe complex object shapes using several primitive colliders on child objects — a compound collider. One Rigidbody on the parent controls the entire physics unit. This is cheaper than MeshCollider and more accurate than a single BoxCollider. For vehicles: separate BoxColliders for the body, bumpers, wheel arches. WheelCollider is a specialized component for realistic suspension behavior; it does not participate in standard OnCollisionEnter events.

What is included in turnkey physics setup

On one recent project, we fixed a character that constantly fell through floors at speeds above 10 m/s. The root cause was a non-convex MeshCollider on the environment and Discrete detection on the character. We replaced the MeshCollider with a set of primitive BoxColliders forming a compound collider, set the character's Rigidbody to Continuous, and adjusted the Layer Matrix to exclude trigger zones from raycast hits. After the fix, the character never fell through, and the CPU overhead increased by only 3%.

Timeline estimates (cost quoted after analysis)

Typical mistakes and how to avoid them

Collider extends beyond the renderer mesh — the character visually passes through a wall but physically hits an invisible barrier. Ensure collider dimensions match the mesh. OnCollisionEnter is not called — one of the objects is a kinematic Rigidbody or a StaticCollider without Rigidbody. OnCollisionEnter requires a Rigidbody on at least one object. For static geometry, OnTriggerEnter on trigger zones is sufficient. Physics objects jitter in place — solver iteration count is too low or the object is under competing forces. Increase Default Solver Iterations from 6 to 10–12 for complex scenes.

Our engineers with years of experience can help you avoid these pitfalls. Order a physics setup — get a consultation and a ready solution.

Gameplay Programming: The Core of Game Mechanics

We often inherit projects with chaotic architecture. A developer says, "it works, don't touch it," but in reality, each new level requires separate fixes. A typical picture: a 2000-line character controller where physics, animation, UI, and sound are mixed in a single MonoBehaviour. Saving via PlayerPrefs with keys like "player_hp_current_value_int". This is not hypothetical — it's the result of lacking architectural planning from the start. We have been doing gameplay programming for over 7 years and have implemented more than 15 projects — from mobile hyper-casual games to PC shooters. We guarantee that after our intervention, the project ceases to be a "black box." Order a free code audit — we'll assess the state in 2 days.

We take such code, audit it, and reorganize it into a modular system. Gameplay programming is the heart of the game. Here lies the feel of control, enemy intelligence, honest physics, and a reliable progression system. If done poorly, no art can save it.

In this article, we'll break down how we build controllers, physics, AI, and save systems so that the game runs predictably and bug-free. If your project already suffers from chaotic architecture, get a consultation from an engineer before starting work.

Character Controller

The character controller sets the tone for the entire game. The first thing a player encounters is control. Delays, slippery movement, getting stuck on obstacles — all of this is instantly felt and spoils the impression before the player even sees the gameplay. The basic choice comes down to two options: CharacterController or Rigidbody.

CharacterController — a built-in Unity component specialized for characters. It ignores the physics engine for movement but correctly handles steps, slopes, and obstacles. Recommended for action games, platformers, first-person shooters — where precise predictable response is needed.

Rigidbody — a physics object. Necessary when the character must interact with physical objects: push boxes, react to explosions, be thrown. Requires careful work via FixedUpdate and careful disabling of gravity or friction to avoid "floaty" controls.

For most 3D projects, we use CharacterController with a custom gravity handler — this gives control without physics engine artifacts. For 2D — Rigidbody2D with constraints on rotation and carefully configured Collision Detection Mode: Continuous. Each decision is made based on genre and target platform.

Physics and Collisions

Rigidbody and colliders are a source of regular problems if not set up correctly from the start. Several rules that save time:

• Collision Detection: Continuous for fast objects (bullets, projectiles) — otherwise they "tunnel" through thin geometry.
• Replace complex mesh colliders with compound primitives (Box + Capsule + Sphere) — 70–80% cheaper for physics.
• Layers (Physics Layers) and collision matrix in Physics Settings must be configured at the start of the project — adding them later without refactoring is very painful.
• All physics calculations go in FixedUpdate, not Update. Otherwise, behavior depends on FPS.

Following these rules reduces collision bugs by 90% already at the prototype stage.

How Does AI Architecture Affect the Game Experience?

Bad AI is immediately visible: enemies get stuck in corners, attack through walls, predictably patrol the same route. The difference between "works" and "works well" is most noticeable here. Let's consider three levels of detail.

State Machines

The most common approach — hierarchical state machine (HSM). Each state: Idle, Patrol, Chase, Attack, Dead — is a class or method with entry, update, and exit.

public enum EnemyState { Idle, Patrol, Chase, Attack, Dead }

private void UpdateStateMachine() {
    switch (_currentState) {
        case EnemyState.Patrol:
            UpdatePatrol();
            if (CanSeePlayer()) TransitionTo(EnemyState.Chase);
            break;
        case EnemyState.Chase:
            _navMeshAgent.SetDestination(_player.position);
            if (InAttackRange()) TransitionTo(EnemyState.Attack);
            if (!CanSeePlayer() && _lostSightTimer > 5f) TransitionTo(EnemyState.Patrol);
            break;
        // ...
    }
}

State Machine works well for enemies with a small number of states (5–8). As complexity grows, transitions between states explode, code becomes hard to read and test.

Behaviour Trees

Behaviour Tree — the next level. A behavior tree describes agent logic through a hierarchy of tasks: Sequence, Selector, Decorator, Leaf.

Advantage over State Machine: each node is atomic and reusable. The CheckLineOfSight node is written once and used in ten trees. Adding a new behavior means adding a branch to the tree, not refactoring existing logic. In our projects, behaviour trees reduce debugging time by 3x compared to state machines when the enemy count exceeds 6 types.

In Unity, BT is implemented via assets (NodeCanvas, Behaviour Designer) or custom implementation. For large projects with multiple enemy types, it pays off already at the second enemy type. Example tree structure for a patrolling enemy:

Root
└── Selector
    ├── Sequence (Combat)
    │   ├── IsPlayerVisible
    │   ├── IsPlayerInRange
    │   └── AttackPlayer
    ├── Sequence (Alert)
    │   ├── HeardSound
    │   └── InvestigatePosition
    └── Sequence (Patrol)
        ├── HasPatrolRoute
        └── FollowPatrolRoute

GOAP — When BT is Not Enough

Goal-Oriented Action Planning — an approach for truly complex AI where the agent must plan a sequence of actions to achieve a goal considering the current world state. Classic example: an enemy that needs to "kill the player." If it has no weapon, it looks for one. If no ammo, it searches for ammo. If the player takes cover, it finds an alternate route. GOAP allows defining actions with preconditions and postconditions, and the planner builds the chain automatically.

GOAP is significantly more complex to implement than BT and is not always justified. For platformers and casual games, it's overkill. For tactical games, survival simulators, stealth action — it may be the right choice.

NavMeshAgent and Navigation

NavMeshAgent — the standard navigation tool in Unity. It works correctly with proper NavMesh and agent settings:

• Agent Radius and Agent Height must exactly match the character's collider.
• Stopping Distance should be tuned to each enemy type's attack range.
• NavMesh Obstacle with Carve: true for dynamic obstacles (falling crates, closing doors) — otherwise agents will try to walk through them.
• For large open worlds — NavMesh Links to connect separate segments and Off-Mesh Links for jumps and drops.

Why Does the Save System Require Versioning?

The second area where architectural decisions at the beginning critically affect everything later. Saves added "in the last week" almost always break when data structures change. Let's consider the tools.

PlayerPrefs — When It Fits and When It Doesn't

PlayerPrefs is a simple key-value store (string, int, float). It is strictly for settings (volume, controls, language). Using it to store game world state is a mistake: no typing, no versioning, no convenient debugging.

JSON Serialization

A working approach for most projects — serialize data to JSON using JsonUtility (built-in, fast, but limited) or Newtonsoft.Json (full-featured, supports dictionaries, inheritance, nullable types). Save system structure:

[Serializable]
public class SaveData {
    public int version = 1;          // versioning
    public PlayerSaveData player;
    public WorldSaveData world;
    public SettingsSaveData settings;
}

public class SaveSystem : MonoBehaviour {
    private const string SAVE_FILE = "/save.json";

    public void Save(SaveData data) {
        string json = JsonConvert.SerializeObject(data, Formatting.Indented);
        File.WriteAllText(Application.persistentDataPath + SAVE_FILE, json);
    }

    public SaveData Load() {
        string path = Application.persistentDataPath + SAVE_FILE;
        if (!File.Exists(path)) return new SaveData();
        string json = File.ReadAllText(path);
        return JsonConvert.DeserializeObject<SaveData>(json);
    }
}

ScriptableObject as Data Container

ScriptableObject — an underused tool for storing game data. Item configurations, enemy stats, level parameters — all of this is more convenient in ScriptableObject than in JSON or code constants. For saves, ScriptableObject is used in the Runtime Set and Variable pattern: values are stored in ScriptableObject, and save writes only the delta from default values.

Save Versioning

The version field at the root of SaveData is not bureaucracy, but necessity. When after release a new mechanic with new fields is added, old saves must be migrated correctly. Migration method:

private SaveData MigrateSaveData(SaveData data) {
    if (data.version < 2) {
        data.player.newField = defaultValue;
        data.version = 2;
    }
    if (data.version < 3) {
        // next migration
        data.version = 3;
    }
    return data;
}

Without versioning, you have to choose between broken saves for players or refusing to change the data structure. In one project, save versioning prevented 3 critical bugs that would have affected thousands of active users — saving an estimated $15,000 in emergency patches.

ScriptableObject Architecture

For medium and large projects, we use an approach popularized by Ryan Hipple's GDC talk Game Architecture with ScriptableObjects.

// Variable-event
[CreateAssetMenu]
public class GameEvent : ScriptableObject {
    private List<GameEventListener> _listeners = new();

    public void Raise() {
        for (int i = _listeners.Count - 1; i >= 0; i--)
            _listeners[i].OnEventRaised();
    }
}

This allows systems in the game to interact without direct references to each other. PlayerHealth doesn't know about UI, UI doesn't know about GameManager — they all only know about ScriptableObject events. The project becomes significantly easier to test and extend. We measured a 4x reduction in coupling compared to direct references.

How We Work: Phases and Results

Each project goes through five phases. Below are indicative timelines and key artifacts. Cost is calculated individually; as a reference, one module (e.g., save system) varies depending on complexity.

Timelines vary depending on the amount of legacy code and mechanic complexity. Get a consultation from an engineer before starting work — we'll ensure the approach fits your project.

What's Included

Upon completion, you receive:

• Architectural documentation for game systems (controller, AI, saves)
• Source code with comments, ready for further development
• Tool settings: Physics Layers, NavMesh, ScriptableObject event project
• Code review of existing modules (if not a from-scratch project)
• Support during integration (2 weeks after handover)

Contact us to discuss details. Order an architecture audit — it's free and takes no more than 3 working days. Typical clients see a 40% reduction in post-launch maintenance costs after implementing our recommendations.