111 lines
8.9 KiB
JavaScript
111 lines
8.9 KiB
JavaScript
/* 原始老师轨迹 → 闭环训练样本（方案 §4.5 / §5）
|
||
* node ml/prep/build_dataset.cjs ml/data/shard_0001.jsonl [outDir=ml/data/samples]
|
||
* 每条 keep=true 的轨迹按 50 Hz 时间律重放，每 3 步取一个样本：
|
||
* X [N,83] 特征（webapp/js/ik-policy.js features()，与部署同一份代码）
|
||
* Y [N,16,6] 未来 16 步每步增量 / (ACT_SCALE·STEP)，夹在 ±1（到终点后补 0）
|
||
* V [N] 剩余时间（s）/ 5
|
||
* SG [N,13] 子目标标签：剩余两个途经点相对当前 q 的差 / 180（不足用终点补）+ 是否需要
|
||
* TID[N] 任务编号（tasks.jsonl 行号所在的全局 id，用于划分与闭环验证）
|
||
* 增广（2026-10-18 首次正式预训练后加大）：40% 样本对 q_t 加 σ=AUG_SIGMA（默认 3°）噪声，标签改为 AUG_BACK（默认 12）步内线性收回老师轨迹；
|
||
* 噪声态须无干涉且间隙 > MIN_GAP_MM，收回途中第 3/6/9/12 步也须无干涉（σ 加大后直接线性拉回可能穿过障碍）。
|
||
* 原为 30%、σ=1°、8 步：闭环一偏出 1° 就是没见过的状态，首次 21850 步闭环到达 0%。更大的偏离由 DAgger 阶段补（ml/dagger）。
|
||
* 已提交前缀长度 r ∈ {0..3} 均匀随机，前缀取老师接下来 r 步的增量。
|
||
* meta [<shard>.meta.json] 打包参数：data_ver、act_scale、增广参数、sig（run_prep.sh 据 sig 判断要不要重新打包）。
|
||
* 2026-10-19（审查 review_data_actions C2）：
|
||
* - 标签按 ACT_SCALE·STEP 归一化（默认 1.25，DATA_VER 2）。原来按 STEP 归一化，巡航时主导轴标签恰好 = ±1、执行端又夹在 ±1，
|
||
* 模型在 ±1 附近的对称误差有一半被截掉，主导轴系统性偏慢；留 25% 余量后老师标签最大 0.8，执行端的 ±1 只防越界。
|
||
* 运行时 inc = clamp(act, ±1)·act_scale·STEP（closed_loop.py / apg / ik-policy.js 同口径，act_scale 经 checkpoint 与 ONNX meta 传递）；
|
||
* - 增广收回标签的亏空结转：被 ±1 截掉的部分记在 prevP 里、下一步补，原来 prevP = 目标点，截掉就丢了（教的是「落后了别追」）；
|
||
* - DAgger 任务带 qPrev / v0（gen_dataset.cjs 写入）：t=0 样本用真实上一帧，时间律从当前速度起步。
|
||
* 特征（83 维）定义不变，FEAT_VER 不变；数据版本由 meta.data_ver 区分，旧打包没有 meta → 训练按 act_scale 1.0 读。 */
|
||
'use strict';
|
||
const fs = require('fs'), path = require('path');
|
||
/* 打包参数（环境变量）与签名：run_prep.sh 用 `node build_dataset.cjs --sig` 取当前签名，与已打包 shard 的 meta.sig 不同就重新打包 */
|
||
const DATA_VER = 2;
|
||
const MIN_GAP_MM = +(process.env.MIN_GAP_MM || 3);
|
||
const AUG_P = +(process.env.AUG_P || 0.4), AUG_SIGMA = +(process.env.AUG_SIGMA || 3), AUG_BACK = +(process.env.AUG_BACK || 12);
|
||
const ACT_SCALE = +(process.env.ACT_SCALE || 1.25);
|
||
const SIG = ['v' + DATA_VER, 's' + ACT_SCALE, 'p' + AUG_P, 'sg' + AUG_SIGMA, 'b' + AUG_BACK, 'g' + MIN_GAP_MM].join('|');
|
||
if (process.argv[2] === '--sig') { console.log(SIG); process.exit(0); }
|
||
const IN = process.argv[2], OUTDIR = process.argv[3] || path.join(__dirname, '../data/samples');
|
||
if (!IN) { console.error('usage: build_dataset.cjs <shard.jsonl> [outDir] | --sig'); process.exit(1); }
|
||
const { S } = require('../../tests/route_bench_setup.cjs')();
|
||
const POL = require('../../webapp/js/ik-policy.js');
|
||
const { timeParam } = require('../gen/timelaw.cjs');
|
||
const { writeNpy } = require('./npy.cjs');
|
||
const { F_DIM, STEP, H } = POL;
|
||
const L = S.LIM, O = S.ORDER;
|
||
const base = path.basename(IN, '.jsonl');
|
||
let st = 0; for (const ch of base) st = (st * 31 + ch.charCodeAt(0)) >>> 0;
|
||
const rnd = () => { st = (st * 1664525 + 1013904223) >>> 0; return st / 4294967296; };
|
||
const gauss = () => Math.sqrt(-2 * Math.log(rnd() + 1e-12)) * Math.cos(2 * Math.PI * rnd());
|
||
const hash = s => { let h = 2166136261; for (const c of s) h = Math.imul(h ^ c.charCodeAt(0), 16777619); return h >>> 0; };
|
||
const X = [], Y = [], V = [], SG = [], TID = [], tasks = [];
|
||
const lines = fs.readFileSync(IN, 'utf8').split('\n').filter(Boolean);
|
||
let nAug = 0, nAugRej = 0, nGapDrop = 0;
|
||
/* 2026-10-18：认证最小间隙 < 3 mm 的老师轨迹不进训练集（MIN_GAP_MM）；噪声增广接受地板同步为 3 mm（原 1 mm）。 */
|
||
const ASTEP = STEP.map(v => v * ACT_SCALE); // 每单位动作对应的度数
|
||
const NT = lines.length, PSTEP = Math.max(1, Math.floor(NT / 10));
|
||
for (let li = 0; li < NT; li++) {
|
||
if (li % PSTEP === 0) console.log(`[prep] ${base} ${li}/${NT}`); // 每 10% 一行，28 个分片并行也看得见进度
|
||
const line = lines[li];
|
||
const r = JSON.parse(line);
|
||
if (!r.keep) continue;
|
||
if (r.cert && r.cert.minGap != null && r.cert.minGap < MIN_GAP_MM) { nGapDrop++; continue; } // 与 stats.cjs 同一过滤
|
||
const tid = hash(r.id) & 0x7fffffff;
|
||
const split = (hash(r.id + '#split') % 100) < 5 ? 'val' : 'train';
|
||
const P = r.path;
|
||
const tp = timeParam(P, { v0: r.v0 || 0 }); // DAgger 起点带速度：时间律从 v0 起步（gen_dataset.cjs 写 v0）
|
||
const traj = tp.traj, taus = tp.taus, N = traj.length - 1;
|
||
tasks.push({ tid, id: r.id, cat: r.cat, mode: r.mode, q0: r.q0, qp0: r.qPrev || null, T: r.T, R: r.R, exec_s: N * POL.DT, travel: r.travel_deg, reached: r.reached, err: r.err, split });
|
||
const vt = tp.cum; // 顶点 τ
|
||
const inter = P.slice(1, -1).map((p, i) => ({ p, tau: vt[i + 1] }));
|
||
for (let t = 0; t <= N; t += 3) {
|
||
let q = traj[t].slice(), qPrev = (t > 0 ? traj[t - 1] : (r.qPrev || traj[0])).slice();
|
||
let off = [0, 0, 0, 0, 0, 0], aug = false;
|
||
if (rnd() < AUG_P) {
|
||
const qn = q.map((v, j) => { let x = v + AUG_SIGMA * gauss(); const t2 = L[O[j]]; if (!t2.cont) x = Math.max(t2.lo, Math.min(t2.hi, x)); return x; });
|
||
nAug++;
|
||
const o2 = qn.map((v, j) => v - q[j]);
|
||
const back = k => traj[Math.min(N, t + k)].map((v, j) => v + o2[j] * Math.max(0, 1 - k / AUG_BACK));
|
||
if (S.collideFree(qn) && S.clearance(qn, 5) > MIN_GAP_MM && [3, 6, 9, 12].every(k => k >= AUG_BACK || S.collideFree(back(k)))) {
|
||
off = o2; q = qn; qPrev = qPrev.map((v, j) => v + off[j]); aug = true;
|
||
} else nAugRej++;
|
||
}
|
||
const pos = k => { const i = Math.min(N, t + k); const w = aug ? Math.max(0, 1 - k / AUG_BACK) : 0; return traj[i].map((v, j) => v + off[j] * w); };
|
||
const y = new Float32Array(H * 6);
|
||
/* prevP = 按标签实际会走到的位置：被 ±1 截掉的亏空留在 prevP 与目标之差里，下一步接着补（2026-10-19，C2） */
|
||
let prevP = q.slice();
|
||
for (let k = 1; k <= H; k++) {
|
||
const pk = pos(k);
|
||
for (let j = 0; j < 6; j++) { const v = Math.max(-1, Math.min(1, (pk[j] - prevP[j]) / ASTEP[j])); y[(k - 1) * 6 + j] = v; prevP[j] += v * ASTEP[j]; }
|
||
}
|
||
const rr = Math.floor(rnd() * 4), prefix = [];
|
||
for (let k = 0; k < rr; k++) { const row = []; for (let j = 0; j < 6; j++) row.push(y[k * 6 + j] * ASTEP[j]); prefix.push(row); }
|
||
const remain = inter.filter(v => v.tau > taus[t] + 0.02);
|
||
const sub = remain.length && rnd() < 0.8 ? remain[0].p : null;
|
||
const gm = S.gapMatrix(q, POL.GAP_CAP);
|
||
const f = POL.features(S, { q, qPrev, T: r.T, R: r.R, prefix, sub }, gm);
|
||
const v1 = remain[0] ? remain[0].p : P[P.length - 1], v2 = remain[1] ? remain[1].p : P[P.length - 1];
|
||
const sg = new Float32Array(13);
|
||
for (let j = 0; j < 6; j++) { sg[j] = (v1[j] - q[j]) / 180; sg[6 + j] = (v2[j] - q[j]) / 180; }
|
||
sg[12] = remain.length ? 1 : 0;
|
||
X.push(f); Y.push(y); V.push((N - t) * POL.DT / 5); SG.push(sg); TID.push(tid);
|
||
}
|
||
}
|
||
fs.mkdirSync(OUTDIR, { recursive: true });
|
||
const n = X.length;
|
||
const cat = (arrs, w) => { const o = new Float32Array(n * w); arrs.forEach((a, i) => o.set(a, i * w)); return o; };
|
||
writeNpy(path.join(OUTDIR, base + '.X.npy'), cat(X, F_DIM), [n, F_DIM]);
|
||
writeNpy(path.join(OUTDIR, base + '.Y.npy'), cat(Y, H * 6), [n, H, 6]);
|
||
writeNpy(path.join(OUTDIR, base + '.V.npy'), Float32Array.from(V), [n]);
|
||
writeNpy(path.join(OUTDIR, base + '.SG.npy'), cat(SG, 13), [n, 13]);
|
||
writeNpy(path.join(OUTDIR, base + '.TID.npy'), TID, [n], 'int32');
|
||
/* meta 在 tasks.jsonl 之前写：tasks.jsonl 仍是「本 shard 打包完成」的标记 */
|
||
fs.writeFileSync(path.join(OUTDIR, base + '.meta.json'), JSON.stringify({ data_ver: DATA_VER, act_scale: ACT_SCALE, aug_p: AUG_P, aug_sigma: AUG_SIGMA,
|
||
aug_back: AUG_BACK, min_gap_mm: MIN_GAP_MM, aug_carry: true, sig: SIG }));
|
||
// 2026-10-18 断点续跑：tasks.jsonl 最后写、先写 .tmp 再改名 —— 它存在就代表这个 shard 的 5 个 npy 都已写完，run_prep.sh 据此跳过
|
||
const tf = path.join(OUTDIR, base + '.tasks.jsonl');
|
||
fs.writeFileSync(tf + '.tmp', tasks.map(t => JSON.stringify(t)).join('\n') + '\n'); fs.renameSync(tf + '.tmp', tf);
|
||
console.log(JSON.stringify({ shard: base, tasks: tasks.length, samples: n, aug: nAug, augRejected: nAugRej, gapDrop: nGapDrop, featVer: POL.FEAT_VER, dataVer: DATA_VER, actScale: ACT_SCALE }));
|