Relax — Goal-Based Form-Finding

Reading mesh…

Presets — saved goal stacks

Geometry

Nodes X18

Nodes Y18

Span X (m)

Span Y (m)

Starting boundary (a bent wire for soap-film)

Diagonal bracing

Goals — the deck

Select

Nothing selected — pick the Select tool: click (⇧ adds), or drag a box. New goals bind to the selection.

Sim

Quasi-static (no bounce — pure relaxation)

Solver quality (rounds)20

Substeps4

Damping (momentum kept)98%

Display

Colour edges by stress Faces Edges Vertices Invert view — compression twin (hang)

Readouts

Import / Export

Import mesh — .3dm / .obj
as the net, or as a static target/collider

Flip to arch on send (Hooke)

Sends the settled geometry to the Truss tool: nodes are flattened to their best-fit vertical plane (the out-of-plane RMS rides along, never hidden) and your anchors become its supports. Truss picks it up automatically for 15 minutes; relax-net.json also downloads as a backup you can import there any time. Geometry and supports only — Relax edge tensions are solver-unit estimates and do not carry over as forces.

Sources & method

The solver (this is how Kangaroo works). The mesh vertices are particles with momentum. Each frame, force goals (gravity, point loads, pressure, wind) nudge the free particles with a damped, displacement-clamped Verlet step. Then every projection goal in your deck computes, for each particle it touches, a target move and a weight (its strength); the solver averages them — position ← Σ(weight·target)/Σ(weight) — and iterates. Competing goals literally negotiate by weight: two soft anchors of strength w and 2w hold a particle at the ⅓ / ⅔ point. Anchored (pinned) points have infinite weight. That weighted-projection scheme is Daniel Piker's Kangaroo2, built on position-based / projective dynamics; the per-step move cap means a hard yank slows down instead of exploding.

The goal library (24). Anchors (full or per-axis) · springs with rest-length modes (measured / slack / zero / mean / absolute), tension-only cables, yield caps · length clamps · equalize · align-to-direction · bending (straighten rods) · hinge (target fold angle) · corner angles · gravity by tributary area · point loads · pressure · one-sided wind · soap film (zero-rest tension = Schek's force density) · Laplacian smoothing (optionally tangential) · quad planarization · on-plane / sphere / line / imported-mesh constraints · floor, sphere-sphere and mesh collisions · target volume. Presets are just saved stacks of these — the four v1 moves carry over unchanged, and every parameter is now yours to mix.

Honest simplifications. Materials are stylised parameter bundles, not engineering values. Wind is a steady push on projected area — no turbulence, no flutter. Soap film and smoothing use uniform edge weights, not the cotangent weights a true discrete minimal surface wants (Pinkall & Polthier 1993). Anchor "reactions" in the readouts are unit-stiffness estimates in solver units — not calibrated Newtons. Hinge and angle goals rotate the wings only (a stated PBD simplification). No external data sources — everything is client-side math on your mesh.

Sources. Piker — Kangaroo / Kangaroo2 (weighted projections); Müller et al. (2007) Position Based Dynamics; Bouaziz et al. (2014) Projective Dynamics; Schek (1974) force density method; Barnes — dynamic relaxation for tension structures; Hooke (1675) — the hanging chain, inverted, stands as an arch (catenary). .3dm read with rhino3dm. (.3dm export is not in this version — export OBJ and open it in Rhino.)

Scope — form-finding only. These are equilibrium shapes under the goals you set. Not a sized structural design — no member sizing, stresses, buckling or code cases. Use it to find a form; an engineer sizes the real one.