From 20ebb39ad26b33ff3b88e5e653582bb4b4d1d69b Mon Sep 17 00:00:00 2001 From: Meta Vens Date: Fri, 11 Sep 2026 21:00:24 +0200 Subject: [PATCH] Give fire control hit authority and plan navigation Replace the aim-error and steering items with a ServerShared fire control system that owns hit resolution and an MPC navigation planner. Co-Authored-By: Claude Opus 5 --- docs/three-gates-scope.md | 67 ++++++++++++++++++++++++++++++++------- 1 file changed, 55 insertions(+), 12 deletions(-) diff --git a/docs/three-gates-scope.md b/docs/three-gates-scope.md index f2c223b5d..2064e9840 100644 --- a/docs/three-gates-scope.md +++ b/docs/three-gates-scope.md @@ -25,23 +25,66 @@ Fight -> loot -> sell and refit at stations -> push to the next gate. 3. **Reason to fight.** Kills already drop gear and cargo (`EntityInstance.cs`). Add Sell and Repair to station services (`TradeMenu` is buy-only; no repair exists). -4. **AI fairness.** Agent steering reads the gravity slope from - `Zone.GetHeight`. Add aim error to `Combat` state lead prediction. Scale - loadout quality by section rather than biasing toward the largest items. -5. **Minimal narrative.** One Ink beat per boss (pre-fight hail, aftermath), +4. **Fire control owns hits.** Manual aiming goes away. The player (and AI) + targets an item on the enemy schematic; a fire control system in + `ServerShared` resolves where shots land, making the existing damage + localization matter. See below. +5. **Navigation planner.** Replaces steering for agent movement. See below. +6. **Minimal narrative.** One Ink beat per boss (pre-fight hail, aftermath), played through the existing `LocalMenu` Ink player. Story placement (`StoryProcessor`, commented out at `Galaxy.cs:239`) stays off. -## Combat experiment +Items 3 and 6 can land wherever there is room; 1 and 2 come first. -Before any combat redesign: prototype a lock-on/turret-only loadout, using the -existing `LockWeapon` and turret behaviors. If aiming disappears and the fight -becomes positioning in gravity wells, heat, shield/weapon energy, and target -choice, the tactical direction is cheap. If it plays too passively, we learn -that before rewriting anything. +## Fire control + +Aiming in space is not fun, and no human aims well enough for the schematic +damage model (armor -> item -> hull per cell) to matter. Fire control moves +skill from aim to decisions: which subsystem to hit, when to commit fire +against heat and visibility budgets, what each weapon group is for. Player and +AI share it, so AI accuracy is no longer superhuman by construction. + +- **Owner:** fire control in `ServerShared`. Decides hit, cell, and damage. +- **Inputs:** target and selected item, sensor/detection quality, range, + target deviation from the predicted intercept, lock time. +- **Outputs:** schematic hits applied by `DamageSchematic`, which moves from + `Gameplay/EntityInstance.cs` into `Entity`. +- **Demoted:** the seven Unity weapon effects (`Projectile`, + `GuidedProjectile`, `Laser`, `ConstantLaser`, `HitscanEffect`, `Lightning`, + `ConstantLightning`) and `HullCollider` become presentation only. Their + `Physics.*` queries no longer decide damage. +- **Projectiles still fly** toward the predicted intercept; the hit is resolved + on arrival against how far the target deviated, so maneuvering is evasion. +- **Deferred Unity-physics surfaces:** ship collision (`HullCollider`), loot + pickup (`ShieldManager`), `TractorBeam`, `Mine`. Not blockers. + +Unity-owned hit resolution was a speed compromise and is the main reason the +simulation cannot run outside Unity. Moving it is not a headless-server goal; +it is putting hit authority where it belongs, which also lets the planner and a +balance harness simulate combat. + +## Navigation planner + +Everything is a gravity well, and steering cannot navigate one. Escaping a +well means pumping energy: rocking (the classic mountain-car problem) or +thrusting tangentially to spiral out. Stealthy movement means riding terrain +and thrusting rarely; thrust adds visibility (`Behaviors/Thruster.cs:83`). + +Approach: sampling-based model-predictive control (MPPI-style). Each agent +rolls out candidate thrust sequences over a cheap model of the real dynamics +(thrust, `Zone.GetHeight` slope, drag from `Ship.cs`) and follows the best. +Cost = goal progress + thrust x visibility + threat exposure + heat. Rocking, +tangential escape, and surfing emerge from one cost; personality is cost +weights. The same rollout can draw the player's predicted coast path. + +Open risks: drag may bleed too much energy for surfing to work; the height +field is time-varying (gas giant ripples, moving bodies), so rollouts must +sample it at future times. Loadout quality scales by section rather than +biasing toward the largest items. ## Out of scope for now Economy simulation, hauling, mining yield, crafting/blueprints, reputation -changes, story placement, multiplayer and `Economy.Server`, and the -CultMesh/daemon rebuild (parked in `F:\Projects\AetheriaEve`). +changes, story placement, multiplayer and `Economy.Server`, running without +Unity as a goal in itself, and the CultMesh/daemon rebuild (parked in +`F:\Projects\AetheriaEve`).