agentleFS
ukanwatukanwat/aaabench/.claude/skills/physics-tuning/SKILL.md

physics-tuning

Skill399 starsChanged 3 months agoMITView on GitHub

Tune game physics for stable, good-feeling motion — fixed vs variable timestep, render interpolation, mass/gravity/drag, continuous collision detection (CCD) to stop tunneling, fixing jitter, and collision layers/masks. Engine-neutral. Use when the user mentions physics feel, jitter, tunneling, fixed timestep, FixedUpdate, CCD, bouncing/unstable physics, or collision layers.

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use when motion jitters, objects pass through walls (tunneling), stacks explode, or movement feels floaty/sticky/laggy.

    line 27
  2. Distinctive:

    Use to decide what goes in the fixed (physics) step vs the render frame, and how to interpolate between them.

    line 29
  3. Distinctive:

    Use to tune gravity, mass, drag, restitution, solver iterations, sleeping, and collision layers/masks.

    line 31

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Physics tuning

Most "bad physics" is not a bug in the engine — it's a mismatch between the fixed-timestep simulation and the variable-rate render loop, or untuned mass/drag/CCD/layer settings. This skill covers the engine-neutral knobs that make physics stable and responsive; pair it with godot-physics or unity-physics for the concrete APIs.

When to use

  • Use when motion jitters, objects pass through walls (tunneling), stacks explode, or movement feels floaty/sticky/laggy.
  • Use to decide what goes in the fixed (physics) step vs the render frame, and how to interpolate between them.
  • Use to tune gravity, mass, drag, restitution, solver iterations, sleeping, and collision layers/masks.

When not to use: for an engine's exact physics nodes/components and collision callbacks, use godot-physics or unity-physics. For movement decisions (when to jump, AI steering) use input-systems and game-ai. For platformer jump-feel specifics like coyote time/jump buffering, that's input/ controller territory — see input-systems and the platformer genre.

Core workflow

  1. Run physics on a fixed timestep. Simulate at a constant rate (e.g. 50–60 Hz). A fixed dt makes the simulation deterministic-ish and stable; a variable dt makes integration and collisions inconsistent.
  2. Put physics work in the physics callback, not the render frame. Apply forces/velocities and read collisions in the fixed step (FixedUpdate / _physics_process), using that step's dt.
  3. Interpolate rendering between physics ticks. The render frame rate ≠ the physics rate, so smoothly interpolate transforms toward the latest physics state, or enable the engine's Rigidbody interpolation, to remove visible stutter.
  4. Tune the body, not the scene. Set mass for relative weight, drag for damping, gravity scale per object, and restitution/friction via materials.
  5. Stop tunneling with CCD on small/fast bodies; cap maximum velocity.
  6. Stabilize stacks/joints with more solver iterations, sane mass ratios, and sleeping for resting bodies.
  7. Verify by feel and stress test. Play at low and high frame rates; throw fast objects at thin walls; stack and shove bodies. Report what you observed.

Patterns

1. Fixed timestep for simulation, render interpolation for smoothness

# Physics callback: runs at the FIXED rate. Use its dt for all integration.
func _physics_process(dt):                  # Unity: void FixedUpdate()
    velocity += gravity * dt                # integrate with the FIXED dt
    move_and_slide()                        # engine resolves collisions this step
    _prev_pos = _curr_pos; _curr_pos = global_position   # record for interpolation

# Render frame: runs as fast as the display. Interpolate between physics states.
func _process(_frame_dt):                   # Unity: void Update()
    var alpha = Engine.get_physics_interpolation_fraction()  # 0..1 within the tick
    visual.global_position = _prev_pos.lerp(_curr_pos, alpha)
# RIGHT: integrate in the fixed step, render via interpolation.
# WRONG: applying forces in _process/Update with frame dt — speed and collisions
# then depend on frame rate and jitter under load.

Most engines offer this for you (Godot physics_interpolation/Rigidbody interpolate; Unity Rigidbody.interpolation = Interpolate). Prefer the built-in before hand-rolling.

2. Stop tunneling: CCD + a speed cap

# Fast, small bodies skip past thin colliders between ticks. Two fixes:
body.continuous_cd = true            # RigidBody3D bool (RigidBody2D: CCD_MODE_* enum). Unity: rb.collisionDetectionMode = Continuous
# Cap velocity so a single step can't move more than ~one collider thickness.
const MAX_SPEED := 40.0
if velocity.length() > MAX_SPEED:
    velocity = velocity.normalized() * MAX_SPEED
# Rule of thumb: max_distance_per_step (= speed / physics_hz) should be < the
# thinnest wall. Raise physics_hz or enable CCD when that fails.

3. Body tuning: mass, drag, gravity scale, material

# Mass is RELATIVE weight in collisions; it does NOT change fall speed (gravity
# accelerates all masses equally). Use drag and gravity_scale to shape feel.
body.mass = 2.0                      # heavier pushes lighter in collisions
body.linear_damp = 0.5               # air drag: higher = stops sooner (Unity: drag)
body.gravity_scale = 1.5             # per-object gravity multiplier (snappier fall)
# Bounce/slide come from the physics material, not code:
material.bounce = 0.2                # restitution 0..1 (Unity: bounciness)
material.friction = 0.8              # surface grip

4. Collision layers and masks (who collides with whom)

# A body is ON its layer(s) and SCANS the layers in its mask. Both directions of a
# pair must be configured for them to interact.
player.collision_layer = LAYER_PLAYER
player.collision_mask  = LAYER_WORLD | LAYER_ENEMY     # player detects world+enemies
pickup.collision_layer = LAYER_PICKUP
pickup.collision_mask  = LAYER_PLAYER                  # pickup only reacts to player
# Unity equivalent: assign GameObject layers and edit the Physics collision matrix
# (or Physics.IgnoreLayerCollision). Keep a named layer constant table, not magic numbers.

Pitfalls

  • Applying forces/movement in the render frame (Update/_process) makes behavior frame-rate dependent — faster PCs run faster, and collisions get flaky. Do simulation in the fixed step.
  • Visible jitter even with a fixed step usually means no render interpolation: the physics rate and display rate beat against each other. Enable interpolation.
  • Tunneling through thin walls: discrete collision misses fast movers. Enable CCD, cap speed, thicken walls, or raise the physics rate.
  • Expecting heavier objects to fall faster. Gravity is acceleration; mass affects collision response, not fall speed. Use gravity_scale/drag for feel.
  • Exploding stacks / jittery joints: mass ratios too extreme, or too few solver iterations. Keep mass ratios modest and raise iteration counts.
  • Bodies that never rest burn CPU and twitch. Enable sleeping and a sensible sleep threshold for resting objects.
  • One-directional layer setup: A's mask includes B but B's mask excludes A. Detection/collision can need both sides; verify the full matrix.
  • Huge dt spikes (load hitches, breakpoints) blow up integration. Clamp the max physics step / substep count so a stall doesn't launch everything.

References

  • references/timestep-and-ccd.md — the fixed-timestep accumulator loop, interpolation math, substepping, CCD modes, solver/iteration tuning, sleeping, and a stability checklist.

Related skills

  • godot-physics, unity-physics — concrete bodies, colliders, and callbacks.
  • input-systems — responsive controls, jump buffering, coyote time.
  • game-ai — agent movement that must agree with the physics step.
  • platformer, fps-shooter — genres whose feel depends on this tuning.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.