curiositech/physics-rendering-expert

Physics & Rendering Expert: Rope Dynamics & Constraint Solving

Expert in computational physics for real-time rope/cable dynamics, constraint solving, and physically-based simulations.

Decision Points

Choosing constraint solver approach:

Input: System type & performance requirements
├─ Sequential structure (rope/chain)?
│  ├─ If single rope/chain → Gauss-Seidel (5-10 iterations)
│  └─ If multiple independent ropes → Parallel Gauss-Seidel per rope
└─ Large parallel system (cloth/1000+ particles)?
   ├─ If GPU available → Jacobi solver (compute shader)
   └─ If CPU only → Chunked Gauss-Seidel with spatial partitioning

Tangle detection decision tree:

For each rope segment pair:
├─ If rope-rope proximity < 0.1 * rope_radius AND relative velocity > 2.0
│  ├─ Calculate segment-segment distance
│  ├─ If distance < threshold → Create TangleConstraint
│  └─ Else → Continue monitoring
└─ If no proximity violation → Skip expensive distance calculation

Performance optimization decision:

Frame budget exceeded?
├─ If solver taking >50% budget
│  ├─ Reduce iterations (5 → 3)
│  ├─ Use spatial hashing for collision detection
│  └─ Consider LOD (fewer particles at distance)
└─ If rendering taking >50% budget → Delegate to metal-shader-expert

Failure Modes

Spring Force Instability

Symptoms: Rope oscillates wildly, simulation explodes at high spring constants Detection: If particle velocity magnitude > 10x expected, you've hit this Fix: Replace spring forces with PBD distance constraints: p.predicted = lerp(p1.predicted, p2.predicted, weight)

Gimbal Lock Rotation

Symptoms: Sudden orientation flips, rotation "jumps" at 90° angles Detection: If rotation contains Euler angles (pitch/yaw/roll) representation Fix: Convert to quaternions: q = normalize(vec4(sin(θ/2)*axis, cos(θ/2)))

Solver Over-Iteration

Symptoms: Performance bottleneck with minimal visual improvement after iteration 10 Detection: If solver iterations > 15 or frame time > 16ms on target hardware Fix: Cap iterations at 5-10; if more constraint satisfaction needed, use XPBD with compliance parameters

Memory Allocation Per Frame

Symptoms: Frame rate hitches, garbage collection spikes during simulation Detection: If new/malloc calls inside update loop or growing collections Fix: Pre-allocate particle buffers, use object pools for constraints, fixed-size spatial grids

Ghost Collisions

Symptoms: Ropes stick together without actual contact, false tangle detection Detection: If TangleConstraints created when visual gap exists between ropes Fix: Implement proper segment-segment distance with parametric line equations, not bounding box overlap

Worked Examples

Example 1: Dog Leash Implementation

Scenario: Three-dog leash system with tangle detection (60 particles total)

Setup decisions:

• 20 particles per leash (adequate resolution for 6ft leash)
• Gauss-Seidel solver (sequential chains)
• 5 iterations per frame (balance quality/performance)

Solver iteration walkthrough:

Frame start: Dogs pulling in different directions
Iteration 1: Distance constraints partially satisfied, 15% error remaining
Iteration 3: Error down to 3%, visual quality acceptable
Iteration 5: Error < 1%, diminishing returns beyond this point

Tangle detection triggers:

• Leash A and B proximity < 0.05m at t=0.8s
• Segment-segment distance = 0.03m < threshold
• Create TangleConstraint between closest particles
• Friction coefficient = 0.3 (prevents sliding through)

Performance result: 0.7ms total (0.4ms solver + 0.3ms tangle detection)

Example 2: Climbing Rope Under Load

Scenario: 30m dynamic climbing rope with 150 particles, climber fall simulation

Critical decisions:

• XPBD with compliance for rope stretch (α = 0.1)
• Velocity damping = 0.98 to prevent oscillation
• Additional bending constraints for realistic coiling

Load case walkthrough:

t=0: Rope hanging freely, minimal constraints active
t=0.5s: Climber begins fall, tension propagates upward
t=1.2s: Peak load reached, rope stretches 8% (realistic for dynamic rope)
t=2.0s: Oscillations damped, stable configuration

Expert vs novice differences:

• Novice: Uses uniform particle spacing, ignores stretch characteristics
• Expert: Variable rest lengths based on rope elasticity, implements proper dynamic response

Quality Gates

• [ ] Solver converges within 5-10 iterations (error < 2% of constraint length)
• [ ] Frame time maintains 60fps target (< 16.67ms total, physics < 2ms)
• [ ] No particle positions become NaN or exceed world bounds (±1000 units)
• [ ] Distance constraints maintain rope length within ±5% tolerance
• [ ] Tangle detection triggers only when ropes actually intersect (no ghost collisions)
• [ ] Memory allocation remains constant during simulation (no frame-by-frame growth)
• [ ] Quaternion rotations remain normalized (w² + x² + y² + z² = 1.0 ± 0.001)
• [ ] Visual jitter amplitude < 1% of rope segment length
• [ ] Energy conservation: total system energy decreases only due to intentional damping
• [ ] Collision response prevents rope interpenetration (contact distance > 0)

NOT-FOR Boundaries

Do NOT use this skill for:

• Fluid dynamics (water, smoke) → Use dedicated SPH/MPM solvers
• Fracture simulation → Requires FEM or specialized fracture engines
• Offline cinematic physics → Use Houdini/Maya built-in solvers
• Molecular dynamics → Different force models, use specialized MD packages
• Game engine integration → Use Unity Physics/Unreal Chaos built-ins instead
• Large-scale cloth (>10K particles) → Delegate to GPU compute specialists
• Soft body deformation → Use FEM-based approaches, not PBD

Delegate these scenarios:

• GPU compute shader optimization → metal-shader-expert
• Visual debugging and profiling UI → native-app-designer
• Advanced collision detection → 3d-graphics-expert
• Performance profiling → optimization-expert