feat(v0.3.1): real survival patterns — biome-aware scout, wedge-relocate, escape-pit-safe

Operator reported the bot wandered the same 50×50 patch for 2 hours
without making any progress toward food. Diagnosis showed three root
causes; this commit addresses all five open improvement_requests
the LLM (postmortem + tuner) flagged automatically.

Research basis (`Voyager`, `Plan4MC`, `GITM`, `Mindcraft`):
  - Coverage / commit-to-cardinal exploration when local scan fails
  - Biome-aware strategy switching using a static affordance table
  - Wedge detector above the skill layer that triggers RELOCATE not
    RETRY (per-skill stuck checks reset on re-entry — useless)
  - Time-in-region bbox heuristic + need-duration AND skill-cycle gate

Concrete changes:

1. `runtime/goal/storyline.js`
   - orient_self.completed: added timeout fallback (HP=full + session
     >120s → done) so barren biomes don't block the bot on step 1.
     Closes improvement #2 'Нет навыка оценки когда сменить район'.
   - first_food.suggestSkill: now picks survive.scout-food (new) when
     no passive mob is nearby; falls back to survive.acquire-food only
     when something is in immediate range.

2. `runtime/biome-affordances.js` (new)
   - Static table: 40+ biomes → {has_passive_mobs, has_trees,
     has_water, has_crops, livable}.
   - Unknown biomes return optimistic defaults to avoid regressions.
   - Closes improvement #1 'Нет навыка целевого поиска еды по биому'.

3. `runtime/skills/scout-food.js` (new — survive.scout-food)
   - Tiered strategy: biome check → scan 32 → scan 64 → commit a
     cardinal for 200 blocks rescanning every 16. On cardinal
     exhaustion, returns code:"exhausted" so the curriculum can
     escalate to village.relocate.
   - In barren biomes (desert/ocean/snowy_plains) the scan is
     SKIPPED — bot walks straight toward the nearest neighbour
     biome that affords passive mobs (8-direction biome probe at
     radius 64).

4. `runtime/awareness/wedge-detector.js` (new)
   - Rolling 10-min position bbox tracker. observe() called every
     tick; isWedged() returns true when bbox<50 AND active need
     unmet >5min AND skill cycles ≥3.
   - markRelocationStarted() suppresses further wedge firings
     until the bot has displaced ≥200b — prevents stack overflow
     of relocate calls.
   - Lives ABOVE the skill layer (in runtime/reflex.js), because
     any per-skill stuck check resets on re-entry.

5. `runtime/skills/relocate.js` (new — village.relocate)
   - 300-block walk in least-recently-used cardinal (per-incident
     memory in ctx.recentRelocations).
   - Re-paths every 32 blocks, soft-tolerates pathfinder failures
     (3 consecutive throws → exit with code:"stuck_in_place").
   - Closes improvement #2 + #4 ('low success rate trigger').

6. `runtime/skills/escape-pit-safe.js` (new — recovery.escape-pit-safe)
   - Surveys 4 cardinals AND ceiling height before committing.
     Picks the direction with most open blocks (≥3, no lava).
     Falls through to pillar-up only if ceiling clear ≥4b. Returns
     code:"no_strategy" if both blocked so curriculum can escalate
     to relocate.
   - Closes improvement #3 'Нет навыка для безопасного выхода'.

7. `runtime/reflex.js`
   - Wedge detector wired before manifesto/storyline. If wedge.wedged
     is true, dispatches village.relocate directly and returns —
     bypasses every other branch.
   - ctx.disableWedge flag for tests.

8. `runtime/coach/advisor-trigger.js`
   - LLM provider outage backoff: 3 consecutive http_400 / timeout /
     network_error → suppress advisor for 10 min. Today's TimeWeb
     gpt-5.4-mini was 400'ing for an hour straight; we were spending
     trigger budget on dead calls. Closes improvement implicit gap
     in #5.

9. `runtime/bot.js`
   - Tracks botSpawnedAt; snapshot._sessionMs exposed for storyline
     orient_self timeout fallback.

Tests: 396 green (was 378, +18):
  - runtime/biome-affordances.test.js — 8 tests
  - runtime/awareness/wedge-detector.test.js — 9 tests
  - runtime/goal/storyline.test.js — 1 new test (orient_self timeout)

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
2026-05-27 23:01:39 +03:00
co-authored by Claude Opus 4.7
parent cf1f895b25
commit 08ac2e1bc1
15 changed files with 1128 additions and 7 deletions
+135
View File
@@ -0,0 +1,135 @@
// Wedge detector — rolling-10min position bounding box.
//
// Sits above the skill layer (per the v0.3.1 design research). The
// reflex calls observe() with the latest position once per tick; on
// every call we evict older entries and recompute the bbox over the
// trailing window. If the bbox stays small for long enough AND a need
// has been unmet for long enough AND the same skill cycled enough
// times, isWedged() returns true and the reflex injects a `relocate`
// task that supersedes the current need until the bot has displaced
// ≥200 blocks from the wedge centre.
//
// This is intentionally NOT inside any single skill — every skill
// resets its own counters when re-entered, so per-skill stuck checks
// can't break a multi-skill cycle.
const WINDOW_MS = 10 * 60 * 1000; // 10 minutes
const MIN_BBOX_FOR_WEDGE = 50; // <50 blocks max-dim → wedge
const MIN_UNMET_NEED_MS = 5 * 60 * 1000; // need unsatisfied for 5+ min
const MIN_SKILL_CYCLES = 3; // same need's skill restarted ≥3x
let _samples = []; // { t, x, z }
let _activeRelocation = null; // { startedAt, fromCenter:{x,z}, headingName }
let _needStartedAt = new Map(); // needId → t when first detected
let _lastNeedId = null;
export function _resetForTest() {
_samples = [];
_activeRelocation = null;
_needStartedAt = new Map();
_lastNeedId = null;
}
/**
* observe({ x, z, now, activeNeedId, recentSkillIds })
*
* Lightweight: called every reflex tick (~1-2s). Updates sliding
* window and need-duration accounting.
*/
export function observe({ x, z, now = Date.now(), activeNeedId = null, recentSkillIds = [] } = {}) {
if (typeof x !== "number" || typeof z !== "number") return;
_samples.push({ t: now, x, z });
// evict
const cutoff = now - WINDOW_MS;
while (_samples.length && _samples[0].t < cutoff) _samples.shift();
// Track need duration
if (activeNeedId !== _lastNeedId) {
_lastNeedId = activeNeedId;
if (activeNeedId && !_needStartedAt.has(activeNeedId)) {
_needStartedAt.set(activeNeedId, now);
}
}
if (activeNeedId && !_needStartedAt.has(activeNeedId)) {
_needStartedAt.set(activeNeedId, now);
}
// If a relocation is in progress, check whether we've travelled
// far enough to clear it.
if (_activeRelocation && typeof _activeRelocation.fromCenter?.x === "number") {
const dx = x - _activeRelocation.fromCenter.x;
const dz = z - _activeRelocation.fromCenter.z;
if (Math.hypot(dx, dz) >= 200) {
_activeRelocation = null;
// Reset all need-duration timers so the post-relocation env
// gets a fair shot at being labelled satisfied / unsatisfied.
_needStartedAt = new Map();
}
}
}
/**
* isWedged({ activeNeedId, recentSkillIds, now })
* → { wedged: true, bboxDim, needAgeMs, skillCycles, centerX, centerZ } | { wedged: false }
*/
export function isWedged({ activeNeedId, recentSkillIds = [], now = Date.now() } = {}) {
if (_activeRelocation) return { wedged: false, reason: "relocating" };
if (_samples.length < 8) return { wedged: false, reason: "insufficient_samples" };
let xs = Infinity, xb = -Infinity, zs = Infinity, zb = -Infinity, cx = 0, cz = 0;
for (const s of _samples) {
if (s.x < xs) xs = s.x;
if (s.x > xb) xb = s.x;
if (s.z < zs) zs = s.z;
if (s.z > zb) zb = s.z;
cx += s.x; cz += s.z;
}
cx /= _samples.length; cz /= _samples.length;
const bboxDim = Math.max(xb - xs, zb - zs);
if (bboxDim >= MIN_BBOX_FOR_WEDGE) return { wedged: false, reason: "bbox_ok", bboxDim };
const needAgeMs = (activeNeedId && _needStartedAt.has(activeNeedId))
? now - _needStartedAt.get(activeNeedId)
: 0;
if (needAgeMs < MIN_UNMET_NEED_MS) return { wedged: false, reason: "need_recent", needAgeMs, bboxDim };
// Skill cycle count: how many distinct dispatches of the same skill
// appear in the recent rolling window.
const skillCycles = countCycles(recentSkillIds);
if (skillCycles < MIN_SKILL_CYCLES) return { wedged: false, reason: "few_cycles", skillCycles, bboxDim };
return { wedged: true, bboxDim, needAgeMs, skillCycles, centerX: cx, centerZ: cz };
}
/**
* markRelocationStarted({ x, z, heading })
* The reflex calls this when it dispatches village.relocate. While
* a relocation is in flight, isWedged() returns false (relocating)
* so we don't fire a SECOND relocation on top.
*/
export function markRelocationStarted({ x, z, heading } = {}) {
_activeRelocation = {
startedAt: Date.now(),
fromCenter: { x, z },
headingName: heading?.name ?? "?",
};
}
export function activeRelocation() { return _activeRelocation; }
function countCycles(ids) {
if (!Array.isArray(ids) || ids.length === 0) return 0;
// A "cycle" = a transition like A → B → A. Count those.
let cycles = 0;
for (let i = 2; i < ids.length; i++) {
if (ids[i] === ids[i - 2] && ids[i] !== ids[i - 1]) cycles++;
if (ids[i] === ids[i - 1] && ids[i - 1] === ids[i - 2]) cycles++;
}
return cycles;
}
// Test exports
export const __testing = {
WINDOW_MS, MIN_BBOX_FOR_WEDGE, MIN_UNMET_NEED_MS, MIN_SKILL_CYCLES,
countCycles,
};
+104
View File
@@ -0,0 +1,104 @@
import { test } from "node:test";
import assert from "node:assert/strict";
import {
observe,
isWedged,
markRelocationStarted,
activeRelocation,
_resetForTest,
__testing,
} from "./wedge-detector.js";
const { WINDOW_MS, MIN_BBOX_FOR_WEDGE, MIN_UNMET_NEED_MS, MIN_SKILL_CYCLES, countCycles } = __testing;
test("countCycles: empty / short → 0", () => {
assert.equal(countCycles([]), 0);
assert.equal(countCycles(["a"]), 0);
assert.equal(countCycles(["a", "b"]), 0);
});
test("countCycles: A→B→A counts as cycle", () => {
assert.equal(countCycles(["a", "b", "a"]), 1);
assert.equal(countCycles(["a", "b", "a", "b", "a"]), 3);
});
test("countCycles: same skill 3+ times in a row also counts", () => {
assert.equal(countCycles(["a", "a", "a"]), 1);
assert.equal(countCycles(["a", "a", "a", "a"]), 2);
});
test("isWedged: insufficient samples → not wedged", () => {
_resetForTest();
const r = isWedged({ activeNeedId: "food" });
assert.equal(r.wedged, false);
assert.equal(r.reason, "insufficient_samples");
});
test("isWedged: large bbox → not wedged", () => {
_resetForTest();
const t0 = 1_000_000_000_000;
// scatter across 200 blocks
for (let i = 0; i < 12; i++) {
observe({ x: i * 30, z: i * 25, now: t0 + i * 1000, activeNeedId: "food", recentSkillIds: [] });
}
const r = isWedged({ activeNeedId: "food", recentSkillIds: ["a", "b", "a", "b"], now: t0 + 13_000 });
assert.equal(r.wedged, false);
assert.equal(r.reason, "bbox_ok");
});
test("isWedged: tight bbox + old need + cycles → WEDGED", () => {
_resetForTest();
const t0 = 1_000_000_000_000;
// 12 samples within a 30-block bbox, spread across ~9 min so they
// stay inside the 10-min sliding window.
for (let i = 0; i < 12; i++) {
observe({
x: 500 + (i % 4) * 8,
z: 500 + Math.floor(i / 4) * 8,
now: t0 + i * 45_000,
activeNeedId: "food",
recentSkillIds: ["acquire-food", "explore.far", "acquire-food", "explore.far", "acquire-food"],
});
}
const r = isWedged({
activeNeedId: "food",
recentSkillIds: ["acquire-food", "explore.far", "acquire-food", "explore.far", "acquire-food", "explore.far"],
now: t0 + 12 * 45_000,
});
assert.equal(r.wedged, true, `expected wedged, got ${JSON.stringify(r)}`);
assert.ok(r.bboxDim < MIN_BBOX_FOR_WEDGE);
assert.ok(r.needAgeMs >= MIN_UNMET_NEED_MS);
assert.ok(r.skillCycles >= MIN_SKILL_CYCLES);
});
test("isWedged: recent need (under threshold) → not wedged", () => {
_resetForTest();
const t0 = 1_000_000_000_000;
for (let i = 0; i < 12; i++) {
observe({ x: 500, z: 500, now: t0 + i * 10_000, activeNeedId: "food" });
}
const r = isWedged({ activeNeedId: "food", recentSkillIds: ["a", "b", "a", "b"], now: t0 + 60_000 });
assert.equal(r.wedged, false);
});
test("markRelocationStarted blocks subsequent wedge for 200b", () => {
_resetForTest();
const t0 = 1_000_000_000_000;
markRelocationStarted({ x: 500, z: 500, heading: { name: "N" } });
assert.ok(activeRelocation());
// Stay tight bbox after relocation start
for (let i = 0; i < 12; i++) {
observe({ x: 510, z: 510, now: t0 + i * 60_000, activeNeedId: "food" });
}
const r = isWedged({ activeNeedId: "food", recentSkillIds: ["a", "b", "a", "b"] });
assert.equal(r.wedged, false);
assert.equal(r.reason, "relocating");
});
test("relocation clears after travelling ≥200 blocks", () => {
_resetForTest();
markRelocationStarted({ x: 0, z: 0, heading: { name: "N" } });
observe({ x: 250, z: 0, activeNeedId: "food" });
assert.equal(activeRelocation(), null, "relocation cleared after 250b displacement");
});