Guide

Game escort mission systems explained

Harbor Siege’s “Convoy of Ash” chapter failed closed beta because the medic NPC blocked narrow catwalks, sprinted into ambush triggers while players looted, and died to a single stray grenade with no revive window. Mission failure rate hit 62% on the first attempt; 38% of retries abandoned at the same choke point. Players did not complain about difficulty — they complained that the escort felt like babysitting a broken robot. The level was well-authored; the missing layer was an escort mission system that defined how the NPC follows, when combat starts, what counts as failure, and how co-op squads share responsibility for the VIP.

Escort missions are among the most polarizing quest types because they combine pathing AI, combat pacing, and narrative stakes in one fragile loop. Done poorly, the player fights the NPC instead of the enemy. Done well, the escort creates readable pressure, teaches map geography, and sells emotional investment in a character who cannot be reduced to a moving objective marker. This guide classifies escort patterns, specifies follow and catch-up behavior, maps threat routing and failure states, walks through Harbor Siege’s convoy refactor, and provides a technique decision table vs static NPCs and fetch-style quests, pitfalls, and a production checklist.

Escort mission taxonomy

Not every “protect the NPC” beat uses the same rules. Classify escorts before wiring AI so engineers and level designers share vocabulary.

Type	Who moves	Typical pacing	Failure mode
Follow escort	Player leads; NPC trails	Exploration, stealth routes	Distance leash, NPC death
Lead escort	NPC sets pace on spline or path	Scripted tours, tutorials	Player falls behind, NPC killed
Convoy / wave escort	Group moves between waypoints	Horde defense, tower segments	Any VIP dies, vehicle destroyed
Bodyguard defense	NPC stationary or slow orbit	Arena survival, extraction timer	HP threshold, time limit
Extract escort	Downed NPC must be carried/dragged	High stakes rescue	Carrier killed, abandon timer

Hybrid missions are common: a follow escort becomes a bodyguard defense at a courtyard, then a lead escort through a finale corridor. Encode phase transitions in the quest graph so follow AI, combat spawners, and UI objectives swap cleanly.

Follow AI: tether, catch-up and doorway etiquette

Follow escorts live or die on movement. Players forgive dumb combat; they rarely forgive an NPC wedged in a doorframe for thirty seconds.

Tether and catch-up

Define three radii around the player anchor (usually the squad leader in co-op):

Comfort radius — NPC walks at match pace; no sprint.
Alert radius — NPC jogs or calls out; UI hint if player sprints away during combat.
Fail radius — optional hard fail, or NPC enters “lost” state and mission pauses with waypoint ping.

Catch-up triggers when distance exceeds alert radius: increase move speed, enable short-path teleport behind the player (with VFX fade) after N seconds stuck, or skip to the next authored wait point. Teleport is invisible to players if rare; abuse it and immersion breaks.

Navmesh and collision

Escort NPCs need a slightly narrower capsule than the player, ghost-through-friendly layers for other escorts, and explicit door protocols: wait at threshold, open door animation, proceed. Test every choke with the escort trailing at 45-degree angles — the failure case players hit. Integrate with your navmesh off-mesh links for jumps and ladders; do not assume the player path is valid for NPCs without validation.

Idle and wait behavior

When the player engages optional content (loot, terminal hack), the escort should enter HoldPosition at the nearest cover node, not wander into line of fire. Cover selection can reuse companion AI cover queries. Vocal bark when ready to move reduces “is she broken?” confusion.

Threat routing: who enemies attack

Escort difficulty is largely threat policy, not raw enemy DPS. Document rules per mission phase:

Player-primary — enemies target players; escort damage only from AoE or flanking. Default for follow escorts; lowest frustration.
Threat table — escort generates aggro proportional to healing or proximity; players must peel. Use in defense phases with clear telegraphs.
Assassin priority — periodic enemies bypass tanks to hit the VIP; requires UI warning and counter tools (interrupt, shield).
Environmental only — escort HP loss from hazards the player must disable; good for puzzle-forward missions.

Sudden threat switches feel unfair. Ramp escort-targeted damage over the first two waves so players learn the rule. In co-op, replicate threat decisions on the host and show escort HP prominently — shared downed/revive rules apply if the escort can be stabilized instead of instant-failing.

Failure states and player agency

Instant mission fail on VIP death maximizes stakes and maximizes rage-quits. Modern escort systems offer graded outcomes:

Policy	Player experience	When to use
Hard fail	Reload checkpoint	Short arenas, horror tone
Stabilize window	Downed escort + revive channel	Co-op shooters, action RPGs
Secondary objective trade	VIP lives, bonus lost	Optional perfection rewards
Narrative branch	Story continues wounded	Single-player RPGs with consequence
Respawn at phase start	Soft reset of segment only	Long multi-phase escorts

Whatever you choose, telegraph it in the briefing UI and reinforce once in the first combat beat. Players who discover hidden fail rules mid-fight blame the game, not their tactics.

Combat pacing along the route

Escort routes should read as a tension graph, not a uniform combat slider.

Quiet beat — dialogue, environmental storytelling, resource pickup. Escort comments on landmarks to sell travel.
Skirmish — small pack; teaches threat routing for this mission. Spawn behind the escort’s forward arc so players face enemies, not chase camera confusion.
Choke defense — narrow geometry; optional deployable cover. Escort holds in a marked safe alcove.
Breather — no spawns; ammo cache; short checkpoint if genre supports it.
Finale wave — escort reaches extraction trigger only after wave clear or simultaneous run.

Trigger volumes should use player-forward conditions (player enters zone AND escort within comfort radius) so speedrunners cannot blaze ahead and leave the VIP in an unloaded combat state. Lead escorts invert this: combat starts when the NPC enters the volume; players must stay inside the fail radius behind them.

Co-op normalization

In squad play, escort AI should anchor to the slowest necessary player or a designated “point” role, not whoever sprinted ahead.

Squad centroid follow — NPC targets midpoint of living players within 20 m; reduces split-party fails.
Shared escort HP — one bar on HUD; damage from any source visible to all.
Ping integration — “Regroup on VIP” ping when escort hits alert radius; ties into waypoint systems.
Role hints — optional overlay: flank / peel / medic when escort enters defense phase.

If one player can drag the escort into failure alone, griefing and frustration follow. Require consensus triggers (two players in extraction zone) or bind escort pace to the party leader flag.

Harbor Siege convoy refactor

After Convoy of Ash metrics, Harbor Siege rebuilt escort infrastructure:

Escort profile asset — per-mission JSON: radii, threat policy, fail mode, catch-up teleport budget, door protocol flag.
Phase FSM — Follow, Hold, Defense, Lead, Extract states with explicit transitions on quest events.
Stabilize window — 12-second downed state on medic escort; co-op channel revive; solo players get one automatic stim per chapter.
Choke redesign — catwalk widened; escort uses side wait markers; ambush trigger gated on escort + full squad in volume.
Threat ramp — first wave player-primary; assassin skirmishers only after wave three with red telegraph VFX.
Telemetry — log stuck time, fail radius breaches, and death cause (environment vs leak vs burst) per attempt.

First-attempt clear rate rose from 38% to 71%; average stuck time on navmesh fell from 14 seconds to under 2 seconds per segment. Player sentiment on escort missions moved from the lowest quest type to mid-pack — not because combat got easier, but because failures felt legible.

Technique decision table

When a full escort system earns its complexity versus simpler patterns.

Goal	Escort system focus	Alternative
Teach map layout with a guide NPC	Lead escort on spline, no combat	VO-only narrator, hologram marker
Emotional stake for a named character	Follow escort + stabilize fail state	Cutscene-only protection
Horde mode pressure	Bodyguard defense + wave spawners	Static generator defend object
Co-op role clarity	Threat table + ping regroup	Generic defend point
Fetch item across the map	Not an escort — use fetch quest	Escort without follow AI (frustrating)
Permanent companion in open world	Companion AI, not mission escort	Phase-gated escort profile reuse

Common pitfalls

NPC sets off triggers early. Gate spawns on player position or squad centroid, not escort pathing alone.
No catch-up after combat. Players loot; escort stands in crossfire and dies to ambient damage.
Doorway and elevator traps. Missing door protocol blocks progression on geometry every player fits through.
Hidden instant fail. VIP dies in one shot without telegraph or stabilize option.
Escort DPS competes with players. NPC deletes enemies; players feel redundant.
Split-party fail in co-op. One runner triggers fail radius for the whole squad without warning.
Identical escort rules for stealth and action. Stealth escorts need noise and detection hooks, not horde spawns.

Production checklist

Document escort type, threat policy, and fail mode in the mission brief.
Author comfort, alert, and fail radii per phase; tune in telemetry.
Implement catch-up (speed, teleport budget, or skip waypoint) and stuck detection.
Test every narrow choke with escort trailing at an angle.
Gate combat triggers on player-forward or squad-centroid conditions.
Surface escort HP and fail rules on HUD before first contact.
Define co-op anchor (leader vs centroid) and regroup pings.
Log death cause and navmesh stuck time per playtest attempt.

Key takeaways

Escort missions are movement systems first. Combat and story fail if follow AI cannot traverse your level.
Threat policy shapes difficulty more than enemy health pools.
Graded failure preserves stakes without cheap reload loops.
Co-op needs a shared anchor so one sprinter does not break the mission for everyone.
Phase FSMs let one mission blend follow, hold, and defense without bespoke scripts per beat.