Game glider and paraglider aerial descent systems explained

Two implementation families: arcade float vs lift-physics glide

Teams conflate “glider” with “feather fall.” They solve different fantasies:

• Arcade float mode — fixed downward speed cap, strong lateral air control, minimal pitch dependency. Mario Galaxy-style leaf, Fortnite umbrella. Skill is routing and timing deploy; physics is forgiving.
• Lift-physics glide — forward speed generates lift that reduces sink; pitching down trades altitude for speed, pitching up trades speed for hang time until stall. Zelda, Genshin, Assassin's Creed. Skill is energy management and reading terrain.

Harbor Cliffs shipped lift-physics for the main canyon (fantasy: “read the wind, thread the terraces”) and arcade float for a tutorial mesa where new players learn deploy without stall anxiety. Mixing both on the same route without UI labeling caused 19% of deaths when veterans pitched up into stall expecting float forgiveness. Label zones or reset physics profile at volume boundaries.

Deploy and glide state machine

Glider feel lives in transitions, not the hang animation. A production FSM typically includes:

• AirborneEligible — minimum height above ground (raycast or clearance timer), not grounded, not in hitstun, glider item unlocked or charge > 0.
• DeployWindup — 8–15 frames of animation; input buffered; horizontal velocity preserved from jump or launch.
• GlideCruise — lift-drag integration active; pitch input maps to angle of attack clamp.
• StallFlutter — speed below v_stall; increased sink, reduced yaw authority; optional auto-recover if player pitches down.
• DiveBoost — pitch past threshold adds forward accel at cost of sink spike; caps at v_max.
• FlareLand — near-ground auto-flare reduces horizontal speed, brief zero-sink window, then walk state.
• Stow — manual cancel returns to plain fall; respect cooldown if abusable in combat.

Harbor's worst bug was deploy eligibility tied to velocity.y < 0 only — players who walked off edges while still slightly rising could not open the glider for 0.4 s, eating the whole terrace margin. Fix: allow deploy when height_above_ground > h_min regardless of vertical velocity sign, or grant coyote-time deploy after leaving ledge (see coyote time patterns).

Lift, drag, and sink-rate tuning

You do not need a flight sim, but you need a consistent energy story. A practical per-tick model:

v_forward = clamp(v_forward + pitch_accel * dt, v_stall, v_max)
lift      = k_lift * v_forward^2 * cos(pitch)
drag      = k_drag * v_forward^2
sink      = g_eff - lift + pitch_sink_bonus
position += forward * v_forward * dt
position.y -= sink * dt

Tuning knobs designers actually touch:

• v_stall — below this, sink spikes; teaches dive-to-recover without instant death.
• v_max — caps dive bombing; pairs with motion blur and camera FOV punch.
• k_lift / k_drag — ratio sets glide ratio (meters forward per meter down); target 4:1 to 8:1 for adventure games.
• pitch limits — clamp nose-up to ~25° so players cannot hover indefinitely without thermals.

Playtest with a debug HUD showing glide ratio and time-to-ground. Harbor tuned from an unrealistic 12:1 (players overshot every landing pad) down to 6.5:1 with two optional thermal columns restoring reach for skilled routes.

Thermals, wind, and environmental coupling

Gliders shine when level volumes rewrite the sink curve:

• Thermal columns — additive upward sink reduction while inside; stack timer cap prevents infinite hover.
• Ridge lift — lateral wind along cliff faces grants bonus lift when flying parallel; rewards terrain reading.
• Downdraft hazards — increase sink and add turbulence jitter; telegraph with particle streaks 1 s ahead.
• Altitude density — optional soft cap: above h_cap, lift coefficient decays so players cannot climb back over the summit gate.

Apply wind as acceleration on velocity, not direct position teleport — same rule as wind zones and tightrope sway. Harbor downdrafts near the waterfall previously snapped players 3 m sideways per frame; force integration cut cliff impacts 44%.

Camera, input, and landing forgiveness

Gliding fails when players cannot judge ground speed or landing angle:

• Chase camera lag — 80–120 ms position lag sells speed; increase lag slightly during dive boost.
• Ground proximity cues — shadow blob scale + altitude beep under 15 m; flare prompt icon at 8 m.
• Landing tolerance volume — expand valid landing capsule on terraces tagged GlideLanding; snap yaw to surface normal under 12° slope.
• Crash vs slide — above v_land_max, trigger roll animation instead of full damage; only hard-crash into NoGlide spikes.
• Analog vs digital pitch — map trigger or stick Y to pitch with deadzone; keyboard gets tap-to-toggle pitch hold.

Harbor added a 0.35 s auto-flare when crossing landing pads — not a snap teleport, but a sink cut and speed bleed that converted many lethal horizontal skids into standing landings.

Harbor Cliffs refactor

After paraglider unlock shipped, descent deaths stayed at 52% until engineers addressed four root causes:

1. Coyote deploy — 12 frames of deploy eligibility after leaving ledge regardless of velocity.y.
2. Lift curve retune — glide ratio 12:1 → 6.5:1; added two thermals for optional cross-canyon route.
3. Wind integration fix — acceleration-based gusts instead of position snaps.
4. Landing pad flare — auto-flare volumes on three terraces; v_land_max raised 18% on grass, unchanged on rock.

Fall deaths on the descent dropped 71% → 18%. Harbor side-quest completion rose 31% because returning via elevator felt optional, not mandatory. The vista remained a reward — the glider made the geography playable instead of decorative.

Technique decision table

Common pitfalls

• Deploy requires falling — edge-walkers cannot open glider in time.
• Constant sink with no forward authority — reads as delayed death, not flight.
• Stall without recovery path — pitching up becomes a trap, not a tradeoff.
• Wind moves position, not velocity — breaks player counter-input and camera.
• Same physics in combat and exploration — stall during boss arena feels broken.
• Landing snap teleport — kills momentum fantasy and hides tuning problems.
• No glide-gate on ascent — players bypass entire climb with one thermal stack.

Designer and engineer checklist

• Pick arcade float vs lift-physics per zone; label or volume-gate profiles.
• Implement DeployWindup → GlideCruise → Stall → FlareLand FSM with coyote deploy off ledges.
• Tune glide ratio (4:1–8:1) with debug HUD; cap v_max and v_stall.
• Apply wind as acceleration; telegraph downdrafts 1 s ahead.
• Author GlideLanding terraces with tolerance capsules and auto-flare.
• Separate crash (spikes) from fast-land roll on grass and sand.
• Add thermal/ridge volumes with stack timers to prevent infinite hover.
• Chase camera with speed-scaled lag; ground shadow for altitude read.
• Gate summit thermals so players cannot re-climb over the intended one-way drop.
• Playtest deploy from run-off, double-jump launch, and zipline dismount handoff.

Key takeaways

• Gliding is energy routing — not a slower gravity constant.
• Deploy windows matter more than wing mesh — coyote deploy fixes edge deaths.
• Stall must be recoverable — dive pitch is the skill loop.
• Landings need authored forgiveness — flare volumes beat invisible perfection.
• Harbor cut descent deaths 71% → 18% with deploy, lift retune, and landing pads — not a shorter canyon.

Related reading

• Wind zones — updrafts, downdrafts, and aerial force fields
• Zipline traversal — authored cable descents with fixed landing
• Grapple hook — point-to-point swings vs sustained glide
• Platformer design — verb economy and difficulty pacing