Lamapi commited on
Commit
85d6e75
·
verified ·
1 Parent(s): 81762c0

Publish Space with valid short description

Browse files
Files changed (14) hide show
  1. .gitattributes +1 -0
  2. README.md +71 -3
  3. app.js +165 -0
  4. benchmark.json +1 -0
  5. core.js +331 -0
  6. evaluation.json +0 -0
  7. index.html +54 -17
  8. model.json +1 -0
  9. nlrt.js +219 -0
  10. nlui.js +123 -0
  11. parity_cases.json +1 -0
  12. preview.png +0 -0
  13. replay.mp4 +3 -0
  14. style.css +53 -26
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ replay.mp4 filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,10 +1,78 @@
1
  ---
2
  title: Next Drive
3
- emoji: 💻
4
  colorFrom: green
5
- colorTo: red
6
  sdk: static
 
7
  pinned: false
 
 
8
  ---
9
 
10
- Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  title: Next Drive
3
+ emoji: 🚦
4
  colorFrom: green
5
+ colorTo: gray
6
  sdk: static
7
+ app_file: index.html
8
  pinned: false
9
+ license: apache-2.0
10
+ short_description: Autonomous driving in simulated intersections
11
  ---
12
 
13
+ # Next Drive
14
+
15
+ Next Drive is a small learned driver for a **simulated** city intersection. Every 0.1 s it chooses one of six
16
+ longitudinal actions (brake hard, brake, ease off, hold speed, speed up, accelerate) while following a straight, left or right route through a traffic
17
+ light, a crosswalk, a lead car that brakes suddenly, oncoming and cross traffic, and parked cars that hide a pedestrian.
18
+
19
+ **Simulation and education only. It is not connected to, and has never been tested on, a real vehicle, and nothing here
20
+ supports a claim about real-road safety.**
21
+
22
+ **Model:** https://huggingface.co/thelamapi/Next-Drive — pick a scene, density and driver; the drive is generated in your browser and can be replayed with its critical events.
23
+
24
+ ## Results on unseen test scenes (450 per driver; seeds ≥ 10000; density 3 never seen in training)
25
+
26
+ | Driver | Collisions | Red-light violations | Goal reached | Hard brakes / drive | Time to goal (s) | Decision ms (mean / p95) |
27
+ |---|---:|---:|---:|---:|---:|---:|
28
+ | Learned + safety check | 4.4% | 0.4% | 89.1% | 1.67 | 21.0 | 0.34 / 0.47 |
29
+ | Learned | 4.4% | 10.0% | 89.8% | 0.95 | 19.3 | 0.28 / 0.40 |
30
+ | Rule-based | 27.1% | 0.0% | 71.1% | 0.95 | 23.3 | 0.02 / 0.07 |
31
+ | Expert (teacher) | 2.7% | 1.3% | 93.6% | 1.01 | 19.2 | 1.61 / 2.38 |
32
+
33
+ Collision / goal-reached rate by scene type:
34
+
35
+ | Scene | Learned + safety | Learned | Rule-based | Expert |
36
+ |---|---:|---:|---:|---:|
37
+ | light | 4% / 96% | 2% / 98% | 1% / 99% | 1% / 99% |
38
+ | crosswalk | 7% / 93% | 1% / 99% | 9% / 86% | 0% / 100% |
39
+ | lead_brake | 8% / 92% | 1% / 99% | 3% / 97% | 2% / 98% |
40
+ | occlusion | 2% / 79% | 10% / 73% | 68% / 32% | 7% / 82% |
41
+ | mixed | 1% / 86% | 8% / 80% | 54% / 42% | 3% / 89% |
42
+
43
+ Collision rate by traffic density:
44
+
45
+ | Density | Learned + safety | Learned | Rule-based | Expert |
46
+ |---|---:|---:|---:|---:|
47
+ | 1 | 1% | 2% | 23% | 0% |
48
+ | 2 | 5% | 3% | 23% | 3% |
49
+ | 3 (unseen) | 7% | 8% | 35% | 5% |
50
+
51
+ Rates count drives with at least one event. Decision latency was measured in Python on NVIDIA L4, single process; the
52
+ Space reports its own browser timing. All 138 failures of the learned drivers are listed in `evaluation.json`
53
+ and can be replayed in the Space.
54
+
55
+ ![Benchmark](preview.png)
56
+
57
+ [Replay video](replay.mp4): one unselected test scene, then failure replays.
58
+
59
+ ## How it works
60
+ * **Simulator:** 1-lane roads, NS/EW signals with yellow and all-red, IDM-like traffic, pedestrians, and line-of-sight occlusion by parked cars. The ego car perceives cars always and pedestrians only when visible within 60 m.
61
+ * **Expert:** evaluates 6×6 two-phase action sequences (1 s + 2 s) at 0.25 s resolution with constant-velocity predictions, the known light schedule, a 2 m + 1.2 s following gap, and a speed cap that allows stopping before occluded gaps.
62
+ * **Learned driver:** MLP 21→64→64→6 on hand-made features (speed, light state, stop-line distance, lead gap, time-to-collision under three accelerations, visible-pedestrian and occlusion distances, route). Behaviour cloning on 300 expert drives (59750 states), then 3 DAgger round(s); iteration 2 was selected on validation seeds.
63
+ * **Safety check:** keeps an action only if braking at −6 m/s² after 0.5 s still avoids every predicted collision and red-light entry; otherwise picks the strongest safe deceleration.
64
+ * **Rule-based baseline:** IDM car following plus stop-line, visible-pedestrian and left-turn-yield rules.
65
+ * The browser runs a line-by-line JavaScript port; Node parity: 8 full drives, 0 mismatches.
66
+
67
+ ## Usage
68
+ ```bash
69
+ python inference.py --seed 10007 --kind mixed --density 2 --driver shield
70
+ python inference.py --selftest
71
+ ```
72
+ `policy.safetensors` holds the PyTorch weights; `model.json` the same weights for NumPy and the browser.
73
+
74
+ ## Limitations
75
+ * Speed-only control on fixed routes; no lane changes, no steering errors, no sensor noise beyond occlusion.
76
+ * Other road users follow simple scripted behaviour and never react to the ego car.
77
+ * The expert is imperfect too (see the table); the learned driver copies its habits, including frequent hard braking at yellow lights.
78
+ * All numbers come from this simulator and say nothing about real driving.
app.js ADDED
@@ -0,0 +1,165 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /* Next Drive Space: generate an intersection scene, let a driver drive it in the browser, replay with events. */
2
+ const { $, h, fmt } = NLUI;
3
+ const DRIVERS = { shield: "Learned + safety check", learned: "Learned driver", rule: "Rule-based driver", expert: "Expert planner (teacher)" };
4
+ const KIND_NAMES = { mixed: "Everything at once", occlusion: "Parked cars hide a pedestrian", crosswalk: "Pedestrians on the crosswalk", lead_brake: "Car ahead brakes suddenly", light: "Traffic light only" };
5
+ let ctx, bench, sc, run, frame = 0, playing = false, last = 0, showHidden = true;
6
+
7
+ function newRun() {
8
+ sc = Core.makeScenario(Number($("seed").value) | 0, $("kind").value, Number($("density").value));
9
+ run = Core.runEpisode(sc, $("driver").value, ctx.layers, true);
10
+ frame = 0; $("scrub").max = run.trace.length - 1;
11
+ summary(); events(); draw();
12
+ return run;
13
+ }
14
+
15
+ function summary() {
16
+ const ms = run.decision_ms.slice().sort((a, b) => a - b);
17
+ const outcome = run.collision ? `collision with ${run.collision_with}` : run.reached ? "reached the goal" : "timed out (45 s)";
18
+ $("outcome").innerHTML = "";
19
+ $("outcome").append(h("div", { class: "kpi" },
20
+ h("div", {}, h("b", { style: `color:${run.collision ? "var(--red)" : run.reached ? "var(--accent)" : "var(--gold)"}` }, outcome), h("span", {}, `route: ${sc.route} · ${fmt(run.time, 1)} s`)),
21
+ h("div", {}, h("b", {}, run.red_violations), h("span", {}, "red-light violations")),
22
+ h("div", {}, h("b", {}, run.hard_brakes), h("span", {}, "hard brakes")),
23
+ h("div", {}, h("b", {}, run.overrides), h("span", {}, "safety overrides")),
24
+ h("div", {}, h("b", {}, fmt(ms.reduce((a, b) => a + b, 0) / ms.length, 2) + " ms"), h("span", {}, `decision latency (p95 ${fmt(ms[Math.floor(0.95 * (ms.length - 1))], 2)})`))));
25
+ }
26
+
27
+ function events() {
28
+ const box = $("events"); box.innerHTML = "";
29
+ if (!run.events.length) box.append(h("span", { class: "small muted" }, "No critical events."));
30
+ for (const e of run.events) {
31
+ const k = Math.max(0, run.trace.findIndex(r => r.t >= e.t - 1e-9));
32
+ box.append(h("button", { class: "chip", style: e.type === "collision" ? "border-color:var(--red);color:var(--red)" : "", onclick: () => { frame = Math.max(0, k - 15); draw(); } },
33
+ `${fmt(e.t, 1)} s · ${e.type.replace("_", " ")}${e.with ? " (" + e.with + ")" : ""}`));
34
+ }
35
+ }
36
+
37
+ function compareAll() {
38
+ const tb = $("cmp"); tb.innerHTML = "";
39
+ tb.append(h("tr", {}, ["Driver", "Outcome", "Time", "Red", "Hard brakes", "ms/decision"].map((x, i) => h("th", { class: i > 1 ? "num" : "" }, x))));
40
+ for (const d of ["shield", "learned", "rule", "expert"]) {
41
+ const r = Core.runEpisode(sc, d, ctx.layers, false);
42
+ const ms = r.decision_ms.reduce((a, b) => a + b, 0) / r.decision_ms.length;
43
+ tb.append(h("tr", {}, h("td", {}, DRIVERS[d]), h("td", { style: r.collision ? "color:var(--red)" : "" }, r.collision ? "collision" : r.reached ? "goal" : "timeout"),
44
+ h("td", { class: "num" }, fmt(r.time, 1)), h("td", { class: "num" }, r.red_violations), h("td", { class: "num" }, r.hard_brakes), h("td", { class: "num" }, fmt(ms, 2))));
45
+ }
46
+ }
47
+
48
+ function draw() {
49
+ const c = $("view"), W = c.clientWidth || 640, H = Math.round(Math.min(W * 0.72, 520)), dpr = window.devicePixelRatio || 1;
50
+ c.width = W * dpr; c.height = H * dpr; c.style.height = H + "px";
51
+ const g = c.getContext("2d"); g.setTransform(dpr, 0, 0, dpr, 0, 0);
52
+ const r = run.trace[Math.min(frame, run.trace.length - 1)];
53
+ const scale = W / 62, cx = Math.max(-40, Math.min(40, r.x + 10 * Math.cos(r.h))), cy = Math.max(-55, Math.min(45, r.y + 10 * Math.sin(r.h)));
54
+ const X = x => W / 2 + (x - cx) * scale, Y = y => H / 2 - (y - cy) * scale;
55
+ g.fillStyle = "#5b8a4a"; g.fillRect(0, 0, W, H); // grass
56
+ g.fillStyle = "#c9c4b8"; g.fillRect(X(-6), 0, 12 * scale, H); g.fillRect(0, Y(6), W, 12 * scale); // sidewalks
57
+ g.fillStyle = "#3a3d42"; g.fillRect(X(-3.5), 0, 7 * scale, H); g.fillRect(0, Y(3.5), W, 7 * scale); // asphalt
58
+ g.strokeStyle = "#f2c94c"; g.lineWidth = 2; g.setLineDash([3 * scale, 3 * scale]);
59
+ g.beginPath(); g.moveTo(X(0), 0); g.lineTo(X(0), Y(3.5)); g.moveTo(X(0), Y(-3.5)); g.lineTo(X(0), H);
60
+ g.moveTo(0, Y(0)); g.lineTo(X(-3.5), Y(0)); g.moveTo(X(3.5), Y(0)); g.lineTo(W, Y(0)); g.stroke(); g.setLineDash([]);
61
+ g.fillStyle = "#eee"; // crosswalk + stop line
62
+ for (let x = -3.2; x < 3.4; x += 0.9) g.fillRect(X(x), Y(-6.0), 0.5 * scale, 3.0 * scale);
63
+ g.fillRect(X(0), Y(-9.9), 3.5 * scale, 0.35 * scale);
64
+ for (const [px, py, pw, ph] of sc.parked) car(g, X, Y, scale, px, py, Math.PI / 2, "#8d939c", pw, ph);
65
+ const goal = Core.pose(sc.route, Core.pathLen(sc.route) - 5); // goal flag
66
+ g.fillStyle = "#0b6e4f"; g.beginPath(); g.arc(X(goal[0]), Y(goal[1]), 0.8 * scale, 0, 7); g.fill();
67
+ g.strokeStyle = "rgba(76,195,154,.85)"; g.lineWidth = 3; g.beginPath(); // driven path
68
+ run.trace.slice(0, frame + 1).forEach((q, k) => (k ? g.lineTo : g.moveTo).call(g, X(q.x), Y(q.y))); g.stroke();
69
+ for (const cr of r.cars) car(g, X, Y, scale, cr[0], cr[1], cr[2], cr[3] === "lead" ? "#7d5ba6" : "#3d6fb6");
70
+ r.peds.forEach(([px, py], k) => {
71
+ const vis = r.visible ? r.visible[k] : true;
72
+ if (!vis && !showHidden) return;
73
+ g.beginPath(); g.arc(X(px), Y(py), 0.45 * scale, 0, 7);
74
+ if (vis) { g.fillStyle = "#f28e2b"; g.fill(); } else { g.setLineDash([3, 3]); g.strokeStyle = "#f28e2b"; g.lineWidth = 2; g.stroke(); g.setLineDash([]); }
75
+ });
76
+ const hit = run.collision && frame >= run.trace.length - 1;
77
+ car(g, X, Y, scale, r.x, r.y, r.h, hit ? "#c0392b" : "#1f9d6b");
78
+ lights(g, X, Y, scale, r.light, r.ew);
79
+ const act = Core.ACTION_NAMES[r.i];
80
+ g.fillStyle = "rgba(20,22,26,.78)"; g.fillRect(8, 8, 238, 62); g.fillStyle = "#fff"; g.font = "600 14px system-ui";
81
+ g.fillText(`${DRIVERS[$("driver").value]}`, 16, 27);
82
+ g.font = "13px system-ui"; g.fillText(`t ${fmt(r.t, 1)} s · ${fmt(r.v * 3.6, 0)} km/h`, 16, 45);
83
+ g.fillStyle = r.override ? "#f0a070" : (r.i <= 1 ? "#ff8a7a" : "#9be3c4"); g.fillText(`decision: ${act}${r.override ? " (safety override)" : ""}`, 16, 62);
84
+ if (hit) { g.fillStyle = "rgba(192,57,43,.9)"; g.font = "700 20px system-ui"; g.fillText(`COLLISION (${run.collision_with})`, W / 2 - 110, 36); }
85
+ $("scrub").value = frame;
86
+ }
87
+
88
+ function car(g, X, Y, s, x, y, hd, col, w, l) {
89
+ w = w || 1.8; l = l || 4.4;
90
+ g.save(); g.translate(X(x), Y(y)); g.rotate(-hd);
91
+ g.fillStyle = col; g.fillRect(-l / 2 * s, -w / 2 * s, l * s, w * s);
92
+ g.fillStyle = "rgba(255,255,255,.55)"; g.fillRect(l * 0.12 * s, -w * 0.38 * s, l * 0.18 * s, w * 0.76 * s);
93
+ g.restore();
94
+ }
95
+
96
+ function lights(g, X, Y, s, ns, ew) {
97
+ const col = k => ["#2ecc71", "#f1c40f", "#e74c3c"][k];
98
+ const box = (x, y, k, label) => {
99
+ g.fillStyle = "#222"; g.fillRect(X(x) - 9, Y(y) - 9, 18, 18);
100
+ g.fillStyle = col(k); g.beginPath(); g.arc(X(x), Y(y), 6, 0, 7); g.fill();
101
+ g.fillStyle = "#fff"; g.font = "10px system-ui"; g.fillText(label, X(x) + 11, Y(y) + 4);
102
+ };
103
+ box(4.8, -11.5, ns, "you"); box(-11.5, -4.8, ew, "cross");
104
+ }
105
+
106
+ function tick(ts) {
107
+ if (playing) {
108
+ if (ts - last > 1000 * 0.1 / Number($("speed").value)) {
109
+ last = ts; frame++;
110
+ if (frame >= run.trace.length - 1) { frame = run.trace.length - 1; playing = false; $("play").textContent = "Replay"; }
111
+ draw();
112
+ }
113
+ }
114
+ requestAnimationFrame(tick);
115
+ }
116
+
117
+ function play() {
118
+ if (!playing && frame >= run.trace.length - 1) frame = 0;
119
+ playing = !playing; $("play").textContent = playing ? "Pause" : "Play";
120
+ }
121
+
122
+ function benchTable() {
123
+ const tb = $("bench"); tb.innerHTML = "";
124
+ tb.append(h("tr", {}, ["Driver", "Collisions", "Red-light", "Reached", "Hard brakes / drive", "ms / decision"].map((x, i) => h("th", { class: i ? "num" : "" }, x))));
125
+ for (const d of ["shield", "learned", "rule", "expert"]) {
126
+ const r = bench.overall[d];
127
+ tb.append(h("tr", {}, h("td", {}, DRIVERS[d]), h("td", { class: "num" }, fmt(100 * r.collision_rate, 1) + "%"), h("td", { class: "num" }, fmt(100 * r.red_rate, 1) + "%"),
128
+ h("td", { class: "num" }, fmt(100 * r.reach_rate, 1) + "%"), h("td", { class: "num" }, fmt(r.hard_brakes, 2)), h("td", { class: "num" }, fmt(r.ms_mean, 2))));
129
+ }
130
+ $("benchnote").textContent = `${bench.episodes_per_driver} unseen test drives per driver (seeds ≥ 10000, densities 1–3; density 3 never seen in training). Timing from Python on the training machine.`;
131
+ const box = $("fails"); box.innerHTML = "";
132
+ for (const f of bench.failures.slice(0, 16)) box.append(h("button", { class: "chip", onclick: () => load(f) }, `${DRIVERS[f.driver]} · ${f.kind} d${f.density} #${f.seed} · ${f.what}`));
133
+ if (!bench.failures.length) box.append(h("span", { class: "small muted" }, "No failures recorded for the learned drivers."));
134
+ }
135
+
136
+ function load(f) {
137
+ $("seed").value = f.seed; $("kind").value = f.kind; $("density").value = f.density; $("driver").value = f.driver;
138
+ newRun(); compareAll(); playing = true; $("play").textContent = "Pause";
139
+ window.scrollTo({ top: 0, behavior: "smooth" });
140
+ }
141
+
142
+ window.__liveTest = function () {
143
+ $("seed").value = "424242"; $("kind").value = "mixed"; $("density").value = "3"; $("driver").value = "shield";
144
+ $("go").click();
145
+ const r = run;
146
+ frame = Math.floor(r.trace.length / 2); draw();
147
+ return { ok: r.steps > 0 && r.trace.length === r.steps && $("cmp").rows.length === 5 && sc.seed === 424242 && sc.density === 3,
148
+ outcome: r.collision ? "collision" : r.reached ? "goal" : "timeout", steps: r.steps, overrides: r.overrides, route: sc.route };
149
+ };
150
+
151
+ NLUI.boot(Core, async c => {
152
+ ctx = c;
153
+ bench = await (await fetch("benchmark.json")).json();
154
+ for (const [k, v] of Object.entries(KIND_NAMES)) $("kind").append(h("option", { value: k }, v));
155
+ for (const [k, v] of Object.entries(DRIVERS)) $("driver").append(h("option", { value: k }, v));
156
+ $("go").addEventListener("click", () => { newRun(); compareAll(); });
157
+ $("rand").addEventListener("click", () => { $("seed").value = 10000 + Math.floor(Math.random() * 89999); newRun(); compareAll(); });
158
+ for (const id of ["kind", "density", "driver"]) $(id).addEventListener("change", () => { newRun(); compareAll(); });
159
+ $("play").addEventListener("click", play);
160
+ $("scrub").addEventListener("input", e => { frame = Number(e.target.value); playing = false; $("play").textContent = "Play"; draw(); });
161
+ $("hidden").addEventListener("change", e => { showHidden = e.target.checked; draw(); });
162
+ window.addEventListener("resize", draw);
163
+ benchTable(); newRun(); compareAll();
164
+ requestAnimationFrame(tick);
165
+ });
benchmark.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"overall": {"shield": {"collision_rate": 0.044444444444444446, "red_rate": 0.0044444444444444444, "reach_rate": 0.8911111111111111, "hard_brakes": 1.6711111111111112, "episodes": 450, "ms_mean": 0.3393771767004915, "ms_p95": 0.46968219995733307, "time_to_goal": 21.018703241895263}, "learned": {"collision_rate": 0.044444444444444446, "red_rate": 0.1, "reach_rate": 0.8977777777777778, "hard_brakes": 0.9466666666666667, "episodes": 450, "ms_mean": 0.2776372892021873, "ms_p95": 0.40007489990330214, "time_to_goal": 19.32029702970297}, "rule": {"collision_rate": 0.27111111111111114, "red_rate": 0.0, "reach_rate": 0.7111111111111111, "hard_brakes": 0.9466666666666667, "episodes": 450, "ms_mean": 0.01625473820577528, "ms_p95": 0.06885959994633593, "time_to_goal": 23.2515625}, "expert": {"collision_rate": 0.02666666666666667, "red_rate": 0.013333333333333334, "reach_rate": 0.9355555555555556, "hard_brakes": 1.011111111111111, "episodes": 450, "ms_mean": 1.6056132565324384, "ms_p95": 2.3759480500530117, "time_to_goal": 19.233729216152017}}, "episodes_per_driver": 450, "failures": [{"driver": "shield", "seed": 10021, "kind": "light", "density": 1, "what": "collision: car"}, {"driver": "shield", "seed": 10181, "kind": "light", "density": 2, "what": "collision: car"}, {"driver": "shield", "seed": 10208, "kind": "light", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10271, "kind": "light", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10368, "kind": "crosswalk", "density": 1, "what": "red light"}, {"driver": "shield", "seed": 10466, "kind": "crosswalk", "density": 1, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10535, "kind": "crosswalk", "density": 2, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10549, "kind": "crosswalk", "density": 2, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10584, "kind": "crosswalk", "density": 2, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10604, "kind": "crosswalk", "density": 3, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10632, "kind": "crosswalk", "density": 3, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 10791, "kind": "lead_brake", "density": 2, "what": "collision: car"}, {"driver": "shield", "seed": 10833, "kind": "lead_brake", "density": 2, "what": "collision: car"}, {"driver": "shield", "seed": 10854, "kind": "lead_brake", "density": 2, "what": "collision: car"}, {"driver": "shield", "seed": 10804, "kind": "lead_brake", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10888, "kind": "lead_brake", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10895, "kind": "lead_brake", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10951, "kind": "lead_brake", "density": 3, "what": "collision: car"}, {"driver": "shield", "seed": 10936, "kind": "occlusion", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 10985, "kind": "occlusion", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 10992, "kind": "occlusion", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 11013, "kind": "occlusion", "density": 1, "what": "red light"}, {"driver": "shield", "seed": 11027, "kind": "occlusion", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 11048, "kind": "occlusion", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 10984, "kind": "occlusion", "density": 2, "what": "timeout"}, {"driver": "shield", "seed": 11047, "kind": "occlusion", "density": 2, "what": "timeout"}, {"driver": "shield", "seed": 11089, "kind": "occlusion", "density": 2, "what": "timeout"}, {"driver": "shield", "seed": 11138, "kind": "occlusion", "density": 2, "what": "timeout"}, {"driver": "shield", "seed": 11159, "kind": "occlusion", "density": 2, "what": "timeout"}, {"driver": "shield", "seed": 11067, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11102, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11144, "kind": "occlusion", "density": 3, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 11179, "kind": "occlusion", "density": 3, "what": "collision: pedestrian"}, {"driver": "shield", "seed": 11228, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11235, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11242, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11249, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11256, "kind": "occlusion", "density": 3, "what": "timeout"}, {"driver": "shield", "seed": 11255, "kind": "mixed", "density": 1, "what": "timeout"}, {"driver": "shield", "seed": 11339, "kind": "mixed", "density": 1, "what": "timeout"}]}
core.js ADDED
@@ -0,0 +1,331 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /* Next Drive compute core: line-by-line port of drive_core.py (simulator, perception, drivers, safety check). */
2
+ (function (root, factory) {
3
+ if (typeof module === "object" && module.exports) module.exports = factory();
4
+ else root.Core = factory();
5
+ })(typeof self !== "undefined" ? self : this, function () {
6
+ "use strict";
7
+ const DT = 0.1, VMAX = 14.0, V_LIMIT = 13.9, T_MAX = 45.0;
8
+ const ACTIONS = [-6.0, -3.0, -1.5, 0.0, 1.0, 2.0];
9
+ const ACTION_NAMES = ["brake hard", "brake", "ease off", "hold speed", "speed up", "accelerate"];
10
+ const LANE = 1.75, S_STOP = 60.0, HALF = 2.2;
11
+ const ROUTES = ["straight", "left", "right"], KINDS = ["light", "crosswalk", "lead_brake", "occlusion", "mixed"];
12
+ const TURN = { right: [-5.75, 4.0], left: [-5.25, 7.0] };
13
+ const ASSETS = ["model.json"];
14
+ const PI = Math.PI;
15
+
16
+ function pathLen(route) { if (route === "straight") return 130.0; const [y0, r] = TURN[route]; return (y0 + 70.0) + r * PI / 2 + 55.0; }
17
+
18
+ function pose(route, s) {
19
+ s = Math.max(0.0, Math.min(s, pathLen(route)));
20
+ if (route === "straight") return [LANE, -70.0 + s, PI / 2];
21
+ const [y0, r] = TURN[route], L1 = y0 + 70.0;
22
+ if (s < L1) return [LANE, -70.0 + s, PI / 2];
23
+ const arc = r * PI / 2;
24
+ if (s < L1 + arc) {
25
+ const th = (s - L1) / r;
26
+ if (route === "right") return [LANE + r - r * Math.cos(th), y0 + r * Math.sin(th), PI / 2 - th];
27
+ return [LANE - r + r * Math.cos(th), y0 + r * Math.sin(th), PI / 2 + th];
28
+ }
29
+ const d = s - L1 - arc;
30
+ if (route === "right") return [LANE + r + d, y0 + r, 0.0];
31
+ return [LANE - r - d, y0 + r, PI];
32
+ }
33
+
34
+ function circles(x, y, h) { const c = Math.cos(h), s = Math.sin(h); return [-1.4, 0.0, 1.4].map(d => [x + d * c, y + d * s]); }
35
+ function hitsCar(e, o, m) { m = m || 0; for (const [ax, ay] of e) for (const [bx, by] of o) { const dx = ax - bx, dy = ay - by, rr = 2.0 + m; if (dx * dx + dy * dy < rr * rr) return true; } return false; }
36
+ function hitsPed(e, px, py, m) { m = m || 0; for (const [ax, ay] of e) { const dx = ax - px, dy = ay - py, rr = 1.35 + m; if (dx * dx + dy * dy < rr * rr) return true; } return false; }
37
+
38
+ function lightNS(sc, t) {
39
+ const G = sc.G, Y = 3.0, R = sc.R, u = (t + sc.phase) % (G + Y + R);
40
+ if (u < G) return [0, G - u];
41
+ if (u < G + Y) return [1, G + Y - u];
42
+ return [2, G + Y + R - u];
43
+ }
44
+ function lightEW(sc, t) {
45
+ const G = sc.G, Y = 3.0, R = sc.R, u = (t + sc.phase) % (G + Y + R);
46
+ if (u < G + Y) return 2;
47
+ const r = u - G - Y;
48
+ if (r < 1.0) return 2; if (r < R - 4.0) return 0; if (r < R - 1.0) return 1; return 2;
49
+ }
50
+
51
+ function makeScenario(seed, kind, density) {
52
+ kind = kind || "mixed"; density = density || 1;
53
+ const rng = NLRT.mulberry32((seed * 7919 + KINDS.indexOf(kind) * 101 + density * 13) % 4294967296);
54
+ const u = rng.random();
55
+ let route = u < 0.5 ? "straight" : (u < 0.75 ? "left" : "right");
56
+ if ((kind === "occlusion" || kind === "crosswalk") && rng.random() < 0.5) route = "straight";
57
+ const G = 7.0 + 6.0 * rng.random(), R = 8.0 + 5.0 * rng.random();
58
+ const sc = { seed, kind, density, route, G, R, phase: rng.random() * (G + 3.0 + R), cars: [], peds: [], parked: [], lead: null };
59
+ for (let k = 0; k < density + 1; k++) for (const d of [1, -1])
60
+ sc.cars.push({ axis: "x", dir: d, lane: -LANE * d, p0: -70.0 * d, v0: 8.0 + 4.0 * rng.random(), t0: k * 7.0 + 5.0 * rng.random() });
61
+ if (route === "left" || kind === "mixed")
62
+ for (let k = 0; k < density; k++) sc.cars.push({ axis: "y", dir: -1, lane: -LANE, p0: 70.0, v0: 8.0 + 4.0 * rng.random(), t0: k * 6.0 + 4.0 * rng.random() });
63
+ if (kind === "crosswalk" || kind === "mixed")
64
+ for (let k = 0; k < density + 1; k++) {
65
+ const d = rng.random() < 0.5 ? 1 : -1;
66
+ sc.peds.push({ x0: -9.0 * d, y0: -7.5 + 1.6 * (rng.random() - 0.5), vx: d * (1.0 + 0.6 * rng.random()), vy: 0.0, t0: 1.0 + 11.0 * rng.random(), x_end: 9.0 * d });
67
+ }
68
+ if (kind === "lead_brake" || kind === "mixed")
69
+ sc.lead = { s0: 14.0 + 10.0 * rng.random(), v0: 9.0 + 3.0 * rng.random(), tb: 2.5 + 5.0 * rng.random(), db: 1.5 + 1.5 * rng.random() };
70
+ if (kind === "occlusion" || kind === "mixed") {
71
+ let y = -46.0; const gaps = [];
72
+ while (y < -16.0) { const L = 4.6; sc.parked.push([4.6, y + L / 2, 2.0, L]); y += L + 0.4 + (rng.random() < 0.45 ? 2.6 : 0.0); gaps.push(y - 0.2); }
73
+ const nEmerge = density;
74
+ for (let k = 0; k < Math.min(nEmerge, gaps.length); k++) {
75
+ const gy = gaps[rng.randint(gaps.length)] - 1.0, D = 12.0 + 10.0 * rng.random();
76
+ sc.peds.push({ x0: 5.8, y0: gy, vx: -(1.2 + 0.4 * rng.random()), vy: 0.0, t0: Math.max(0.0, (gy + 70.0 - HALF - D) / 10.0), x_end: -6.0 });
77
+ }
78
+ }
79
+ return sc;
80
+ }
81
+
82
+ function initWorld(sc) {
83
+ const w = { t: 0.0, s: 0.0, v: 10.0, a: 0.0, cars: [], peds: [], lead: null };
84
+ for (const c of sc.cars) w.cars.push({ p: c.p0, v: c.v0, active: false, done: false, axis: c.axis, dir: c.dir, lane: c.lane, v0: c.v0, t0: c.t0 });
85
+ for (const p of sc.peds) w.peds.push({ x: p.x0, y: p.y0, vx: 0.0, vy: 0.0, done: false, t0: p.t0, x_end: p.x_end, svx: p.vx });
86
+ if (sc.lead) w.lead = { s: sc.lead.s0, v: sc.lead.v0, ...sc.lead };
87
+ return w;
88
+ }
89
+
90
+ function carXYH(c) {
91
+ if (c.axis === "x") return [c.p, c.lane, c.dir > 0 ? 0.0 : PI];
92
+ return [c.lane, c.p, c.dir > 0 ? PI / 2 : -PI / 2];
93
+ }
94
+
95
+ function stepAgents(sc, w) {
96
+ const t = w.t, ew = lightEW(sc, t), ns = lightNS(sc, t)[0];
97
+ const lanes = new Map();
98
+ for (const c of w.cars) {
99
+ if (!c.active && !c.done && t >= c.t0) c.active = true;
100
+ if (c.active) { const k = c.axis + c.dir; if (!lanes.has(k)) lanes.set(k, []); lanes.get(k).push(c); }
101
+ }
102
+ for (const cs of lanes.values()) {
103
+ cs.sort((a, b) => (-a.p * a.dir) - (-b.p * b.dir));
104
+ cs.forEach((c, i) => {
105
+ const light = c.axis === "x" ? ew : ns, front = c.p * c.dir + HALF, dStop = -10.0 - front;
106
+ let a = Math.min(2.0, 1.5 * (c.v0 - c.v));
107
+ if (light !== 0 && dStop > -1.0 && c.v * c.v / (2 * 3.0) < dStop + 1.0) a = Math.min(a, -c.v * c.v / (2 * Math.max(dStop - 0.5, 0.3)));
108
+ if (i > 0) { const gap = (cs[i - 1].p - c.p) * c.dir - 4.4; if (gap < 6.0 + c.v * 1.2) a = Math.min(a, gap > 3.0 ? -3.0 : -6.0); }
109
+ c.v = Math.max(0.0, c.v + Math.max(-6.0, a) * DT);
110
+ c.p += c.dir * c.v * DT;
111
+ if (c.p * c.dir > 75.0) { c.active = false; c.done = true; }
112
+ });
113
+ }
114
+ for (const p of w.peds) {
115
+ if (p.done) continue;
116
+ if (t >= p.t0) { p.vx = p.svx; p.x += p.vx * DT; if ((p.x - p.x_end) * p.svx > 0) p.done = true; }
117
+ }
118
+ const L = w.lead;
119
+ if (L !== null) {
120
+ let a;
121
+ if (L.tb <= t && t < L.tb + L.db) a = -6.0;
122
+ else {
123
+ a = Math.min(2.0, 1.5 * (L.v0 - L.v));
124
+ const dStop = S_STOP - (L.s + HALF);
125
+ if (ns !== 0 && dStop > -1.0 && L.v * L.v / (2 * 3.0) < dStop + 1.0) a = Math.min(a, -L.v * L.v / (2 * Math.max(dStop - 0.5, 0.3)));
126
+ }
127
+ L.v = Math.max(0.0, L.v + Math.max(-6.0, a) * DT);
128
+ L.s += L.v * DT;
129
+ }
130
+ }
131
+
132
+ function segHitsRect(x0, y0, x1, y1, rect) {
133
+ const [cx, cy, wd, ht] = rect, xmin = cx - wd / 2, xmax = cx + wd / 2, ymin = cy - ht / 2, ymax = cy + ht / 2;
134
+ let t0 = 0.0, t1 = 1.0;
135
+ for (const [p, q] of [[-(x1 - x0), x0 - xmin], [x1 - x0, xmax - x0], [-(y1 - y0), y0 - ymin], [y1 - y0, ymax - y0]]) {
136
+ if (p === 0) { if (q < 0) return false; }
137
+ else {
138
+ const r = q / p;
139
+ if (p < 0) { if (r > t1) return false; if (r > t0) t0 = r; }
140
+ else { if (r < t0) return false; if (r < t1) t1 = r; }
141
+ }
142
+ }
143
+ return true;
144
+ }
145
+
146
+ function observe(sc, w) {
147
+ const [ex, ey, eh] = pose(sc.route, w.s), eye = [ex + 1.0 * Math.cos(eh), ey + 1.0 * Math.sin(eh)];
148
+ const agents = [];
149
+ for (const c of w.cars) if (c.active) { const [x, y, h] = carXYH(c); agents.push({ kind: "car", x, y, h, vx: c.v * Math.cos(h), vy: c.v * Math.sin(h) }); }
150
+ const L = w.lead;
151
+ if (L !== null && L.s < pathLen(sc.route)) { const [x, y, h] = pose(sc.route, L.s); agents.push({ kind: "lead", x, y, h, vx: L.v * Math.cos(h), vy: L.v * Math.sin(h), s: L.s, v: L.v }); }
152
+ const visible = [];
153
+ for (const p of w.peds) {
154
+ if (p.done) continue;
155
+ const dx = p.x - eye[0], dy = p.y - eye[1];
156
+ const vis = dx * dx + dy * dy < 3600.0 && !sc.parked.some(r => segHitsRect(eye[0], eye[1], p.x, p.y, r));
157
+ visible.push(vis);
158
+ if (vis) agents.push({ kind: "ped", x: p.x, y: p.y, h: 0.0, vx: p.vx, vy: p.vy });
159
+ }
160
+ const phantoms = [], sFront = w.s + HALF;
161
+ for (const r of sc.parked) {
162
+ const gy = r[1] + r[3] / 2 + 0.3, ahead = (gy + 70.0) - sFront;
163
+ if (0.0 < ahead && ahead < 35.0 && segHitsRect(eye[0], eye[1], 4.2, gy, r)) phantoms.push({ x: 4.2, y: gy, vx: -1.4, vy: 0.0, ahead });
164
+ }
165
+ const [ls, rem] = lightNS(sc, w.t);
166
+ return { t: w.t, s: w.s, v: w.v, a: w.a, route: sc.route, agents, phantoms, light: ls, remaining: rem, sc, visible };
167
+ }
168
+
169
+ function rolloutHits(obs, plan, horizon, dt, mCar, mPed) {
170
+ horizon = horizon == null ? 3.0 : horizon; dt = dt == null ? 0.25 : dt; mCar = mCar == null ? 0.4 : mCar; mPed = mPed == null ? 0.6 : mPed;
171
+ let s = obs.s, v = obs.v; const route = obs.route, n = Math.round(horizon / dt);
172
+ let excess = 0.0, redT = null;
173
+ const lead = obs.agents.filter(a => a.kind === "lead");
174
+ for (let k = 1; k <= n; k++) {
175
+ const tau = k * dt, a = plan(k, tau);
176
+ v = Math.min(VMAX, Math.max(0.0, v + a * dt));
177
+ const sPrev = s; s = s + v * dt;
178
+ if (v > V_LIMIT) excess += v - V_LIMIT;
179
+ for (const ph of obs.phantoms) { const d = ph.y + 70.0 - (s + HALF); if (d > -1.0) excess += Math.max(0.0, v - (Math.sqrt(2 * 4.0 * Math.max(d - 2.0, 0.0)) + 3.0)); }
180
+ for (const ag of lead) { const gap = ag.s + ag.v * tau - s - 4.4; excess += 0.5 * Math.max(0.0, 2.0 + 1.2 * v - gap); }
181
+ if (redT === null && s > sPrev) {
182
+ const lt = lightNS(obs.sc, obs.t + tau)[0];
183
+ const crossing = sPrev + HALF < S_STOP - 0.5 && S_STOP - 0.5 <= s + HALF;
184
+ const creeping = S_STOP - 0.5 <= obs.s + HALF && obs.s + HALF < S_STOP + 6.0 && obs.v < 3.0;
185
+ if (lt === 2 && (crossing || creeping)) redT = tau;
186
+ }
187
+ const [ex, ey, eh] = pose(route, s), ec = circles(ex, ey, eh);
188
+ for (const ag of obs.agents) {
189
+ const px = ag.x + ag.vx * tau, py = ag.y + ag.vy * tau;
190
+ if ((px - ex) * (px - ex) + (py - ey) * (py - ey) > 64.0) continue;
191
+ if (ag.kind === "ped") { if (hitsPed(ec, px, py, mPed)) return [tau, redT, s, excess]; }
192
+ else if (hitsCar(ec, circles(px, py, ag.h), mCar)) return [tau, redT, s, excess];
193
+ }
194
+ }
195
+ return [null, redT, s, excess];
196
+ }
197
+
198
+ function expert(obs) {
199
+ let best = 0, bestCost = Infinity;
200
+ ACTIONS.forEach((a1, i) => {
201
+ for (const a2 of ACTIONS) {
202
+ const [hit, red, sEnd, excess] = rolloutHits(obs, (k, tau) => (tau <= 1.0 + 1e-9 ? a1 : a2));
203
+ let cost = 0.0;
204
+ if (hit !== null) cost += 1e5 - 1e4 * hit;
205
+ if (red !== null) cost += 2e4;
206
+ cost += -10.0 * (sEnd - obs.s);
207
+ cost += a1 === -6.0 ? 40.0 : (a1 === -3.0 ? 8.0 : 0.0);
208
+ cost += 3.0 * Math.abs(a1 - obs.a);
209
+ cost += 50.0 * excess;
210
+ if (cost < bestCost) { best = i; bestCost = cost; }
211
+ }
212
+ });
213
+ return best;
214
+ }
215
+
216
+ function ruleDriver(obs) {
217
+ const v = obs.v, s = obs.s, route = obs.route, sFront = s + HALF, obst = [];
218
+ for (const ag of obs.agents) if (ag.kind === "lead") obst.push([ag.s - s - 4.4, ag.v]);
219
+ const dStop = S_STOP - sFront;
220
+ if (dStop > 0 && (obs.light === 2 || (obs.light === 1 && v * v / (2 * 3.0) < dStop))) obst.push([dStop - 0.5, 0.0]);
221
+ for (const ag of obs.agents) {
222
+ if (ag.kind !== "ped") continue;
223
+ for (let j = 0; j < 51; j += 2) {
224
+ const [px, py] = pose(route, s + j), fx = ag.x + ag.vx * 1.5, fy = ag.y + ag.vy * 1.5;
225
+ const d1 = (ag.x - px) * (ag.x - px) + (ag.y - py) * (ag.y - py), d2 = (fx - px) * (fx - px) + (fy - py) * (fy - py);
226
+ if (Math.min(d1, d2) < 6.76) { obst.push([j - HALF - 1.5, 0.0]); break; }
227
+ }
228
+ }
229
+ if (route === "left") {
230
+ const y0 = TURN.left[0], dTurn = (y0 + 70.0) - sFront;
231
+ if (dTurn > -1.0) for (const ag of obs.agents) if (ag.kind === "car" && ag.vy < -0.5 && -4.0 < ag.y && ag.y < 45.0 && ag.x < 0) { obst.push([dTurn - 0.5, 0.0]); break; }
232
+ }
233
+ const q = v / 12.0, q4 = (q * q) * (q * q);
234
+ let a = 1.5 * (1.0 - q4);
235
+ for (let [gap, vo] of obst) {
236
+ gap = Math.max(gap, 0.1);
237
+ const sstar = 2.0 + v * 1.2 + v * (v - vo) / (2 * Math.sqrt(1.5 * 3.0));
238
+ a = Math.min(a, 1.5 * (1.0 - q4 - (sstar / gap) * (sstar / gap)));
239
+ }
240
+ a = Math.max(-6.0, Math.min(2.0, a));
241
+ let bi = 0; for (let i = 1; i < ACTIONS.length; i++) if (Math.abs(ACTIONS[i] - a) < Math.abs(ACTIONS[bi] - a)) bi = i;
242
+ return bi;
243
+ }
244
+
245
+ function features(obs) {
246
+ const v = obs.v, s = obs.s, dStop = S_STOP - (s + HALF);
247
+ const f = [v / 14.0, obs.a / 6.0, Math.min(Math.max(dStop, 0.0), 60.0) / 60.0, dStop < 0 ? 1.0 : 0.0,
248
+ obs.light === 0 ? 1.0 : 0.0, obs.light === 1 ? 1.0 : 0.0, obs.light === 2 ? 1.0 : 0.0, Math.min(obs.remaining, 15.0) / 15.0];
249
+ const lead = obs.agents.filter(a => a.kind === "lead");
250
+ if (lead.length) f.push(Math.min(Math.max(lead[0].s - s - 4.4, 0.0), 50.0) / 50.0, Math.max(-1.0, Math.min(1.0, (lead[0].v - v) / 14.0)));
251
+ else f.push(1.0, 0.0);
252
+ for (const acc of [0.0, -3.0, 1.0]) { const hit = rolloutHits(obs, () => acc, 5.0)[0]; f.push((hit !== null ? hit : 5.0) / 5.0); }
253
+ const [ex, ey] = pose(obs.route, s);
254
+ const peds = obs.agents.filter(a => a.kind === "ped").map(a => Math.sqrt((a.x - ex) * (a.x - ex) + (a.y - ey) * (a.y - ey)));
255
+ f.push(peds.length ? Math.min(Math.min(...peds), 40.0) / 40.0 : 1.0);
256
+ f.push(obs.phantoms.length ? Math.min(Math.min(...obs.phantoms.map(p => p.ahead)), 40.0) / 40.0 : 1.0);
257
+ f.push(obs.agents.some(a => a.kind === "car" && Math.abs(a.x) < 12.0 && Math.abs(a.y) < 12.0) ? 1.0 : 0.0);
258
+ for (const r of ROUTES) f.push(obs.route === r ? 1.0 : 0.0);
259
+ f.push(Math.min(v * v / 12.0, 60.0) / 60.0);
260
+ f.push(obs.light !== 0 && dStop > 0 && v * v / 6.0 < dStop + 1.0 ? 1.0 : 0.0);
261
+ return f;
262
+ }
263
+
264
+ function learned(obs, layers) { return NLRT.argmax(NLRT.forward(layers, NLRT.T([21], Float64Array.from(features(obs)))).data); }
265
+
266
+ function shield(obs, i) {
267
+ const unsafe = a => {
268
+ const [hit, red] = rolloutHits(obs, (k, tau) => (tau <= 0.5 + 1e-9 ? a : -6.0), 3.0, 0.25, 0.2, 0.4);
269
+ if (hit !== null) return true;
270
+ if (red !== null) return rolloutHits(obs, () => -6.0, 3.0, 0.25, 0.2, 0.4)[1] === null;
271
+ return false;
272
+ };
273
+ if (!unsafe(ACTIONS[i])) return [i, false];
274
+ for (let j = i - 1; j >= 0; j--) if (!unsafe(ACTIONS[j])) return [j, true];
275
+ return [0, true];
276
+ }
277
+
278
+ function runEpisode(sc, driver, layers, record) {
279
+ record = record !== false;
280
+ const w = initWorld(sc), route = sc.route, goal = pathLen(route) - 5.0;
281
+ const out = { collision: false, collision_with: null, red_violations: 0, reached: false, hard_brakes: 0, overrides: 0, steps: 0, time: 0.0,
282
+ decision_ms: [], trace: [], events: [], actions: [] };
283
+ let prevA = 0.0;
284
+ const now = () => (typeof performance !== "undefined" ? performance.now() : Date.now());
285
+ for (let step = 0; step < Math.round(T_MAX / DT); step++) {
286
+ const obs = observe(sc, w);
287
+ const t0 = now(); let i, override = false;
288
+ if (driver === "expert") i = expert(obs);
289
+ else if (driver === "rule") i = ruleDriver(obs);
290
+ else if (driver === "learned") i = learned(obs, layers);
291
+ else [i, override] = shield(obs, learned(obs, layers));
292
+ out.decision_ms.push(now() - t0);
293
+ const a = ACTIONS[i];
294
+ out.actions.push(i);
295
+ if (override) out.overrides++;
296
+ if (a <= -6.0 && prevA > -6.0) { out.hard_brakes++; out.events.push({ t: Math.round(w.t * 100) / 100, type: "hard_brake" }); }
297
+ prevA = a;
298
+ const sFrontPrev = w.s + HALF;
299
+ w.v = Math.min(VMAX, Math.max(0.0, w.v + a * DT));
300
+ w.s += w.v * DT; w.a = a;
301
+ stepAgents(sc, w);
302
+ w.t = (step + 1) * DT;
303
+ if (sFrontPrev < S_STOP && S_STOP <= w.s + HALF && lightNS(sc, w.t)[0] === 2) { out.red_violations++; out.events.push({ t: Math.round(w.t * 100) / 100, type: "red_light" }); }
304
+ const [ex, ey, eh] = pose(route, w.s), ec = circles(ex, ey, eh);
305
+ let hit = null;
306
+ for (const c of w.cars) if (c.active && hitsCar(ec, circles(...carXYH(c)))) hit = "car";
307
+ if (w.lead !== null && w.lead.s < pathLen(route) && hitsCar(ec, circles(...pose(route, w.lead.s)))) hit = "lead car";
308
+ for (const p of w.peds) if (!p.done && hitsPed(ec, p.x, p.y)) hit = "pedestrian";
309
+ if (record) {
310
+ out.trace.push({ t: w.t, x: ex, y: ey, h: eh, v: w.v, i, override,
311
+ cars: w.cars.filter(c => c.active).map(carXYH).concat(w.lead !== null && w.lead.s < pathLen(route) ? [pose(route, w.lead.s).concat(["lead"])] : []),
312
+ peds: w.peds.filter(p => !p.done).map(p => [p.x, p.y]), visible: obs.visible, light: lightNS(sc, w.t)[0], ew: lightEW(sc, w.t) });
313
+ }
314
+ out.steps = step + 1; out.time = w.t;
315
+ if (hit) { out.collision = true; out.collision_with = hit; out.events.push({ t: Math.round(w.t * 100) / 100, type: "collision", with: hit }); break; }
316
+ if (w.s >= goal) { out.reached = true; break; }
317
+ }
318
+ return out;
319
+ }
320
+
321
+ function init(assets) { const m = NLRT.decode(assets["model.json"]); return { layers: m.nets.policy, meta: m.meta }; }
322
+
323
+ function runCase(ctx, input) {
324
+ const sc = makeScenario(input.seed, input.kind, input.density);
325
+ const r = runEpisode(sc, input.driver, ctx.layers, false);
326
+ return { route: sc.route, collision: r.collision, red_violations: r.red_violations, reached: r.reached, hard_brakes: r.hard_brakes,
327
+ overrides: r.overrides, steps: r.steps, actions_hash: r.actions.reduce((h, a, k) => (h * 31 + a + k) % 1000000007, 7) };
328
+ }
329
+
330
+ return { ASSETS, init, ACTIONS, ACTION_NAMES, KINDS, ROUTES, S_STOP, HALF, LANE, makeScenario, runEpisode, pose, pathLen, lightNS, lightEW, features, runCase };
331
+ });
evaluation.json ADDED
The diff for this file is too large to render. See raw diff
 
index.html CHANGED
@@ -1,19 +1,56 @@
1
  <!doctype html>
2
- <html>
3
- <head>
4
- <meta charset="utf-8" />
5
- <meta name="viewport" content="width=device-width" />
6
- <title>My static Space</title>
7
- <link rel="stylesheet" href="style.css" />
8
- </head>
9
- <body>
10
- <div class="card">
11
- <h1>Welcome to your static Space!</h1>
12
- <p>You can modify this app directly by editing <i>index.html</i> in the Files and versions tab.</p>
13
- <p>
14
- Also don't forget to check the
15
- <a href="https://huggingface.co/docs/hub/spaces" target="_blank">Spaces documentation</a>.
16
- </p>
17
- </div>
18
- </body>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  </html>
 
1
  <!doctype html>
2
+ <html lang="en">
3
+ <head>
4
+ <meta charset="utf-8">
5
+ <meta name="viewport" content="width=device-width, initial-scale=1">
6
+ <link rel="icon" href="data:,">
7
+ <title>Next Drive</title>
8
+ <link rel="stylesheet" href="style.css">
9
+ </head>
10
+ <body>
11
+ <main>
12
+ <header>
13
+ <h1>Next Drive</h1>
14
+ <p>A small learned driver crosses a city intersection: traffic lights, people on the crosswalk, a car ahead that slams on the brakes, and parked cars hiding a pedestrian. Pick a scene and a driver; the drive is generated in your browser, decision by decision.</p>
15
+ <div class="badges"><span class="badge warn">Simulation and education only · not for real vehicles</span><span class="badge">Live in-browser driving</span><a class="badge" href="https://huggingface.co/thelamapi/Next-Drive">Model repo</a></div>
16
+ </header>
17
+ <section class="card">
18
+ <div class="row">
19
+ <select id="kind" style="flex:2;min-width:180px" aria-label="scenario"></select>
20
+ <select id="density" style="flex:1;min-width:110px" aria-label="traffic density"><option value="1">light traffic</option><option value="2" selected>busy</option><option value="3">very busy (unseen)</option></select>
21
+ <select id="driver" style="flex:2;min-width:180px" aria-label="driver"></select>
22
+ </div>
23
+ <div class="row" style="margin-top:8px">
24
+ <label for="seed" style="margin:0">Scene seed</label><input type="text" id="seed" value="10007" inputmode="numeric" style="width:110px">
25
+ <button id="go">Drive</button><button id="rand" class="ghost">New random scene</button>
26
+ <label style="margin:0"><input type="checkbox" id="hidden" checked> show pedestrians the car cannot see (dashed)</label>
27
+ </div>
28
+ <canvas id="view" style="margin-top:10px"></canvas>
29
+ <div class="row" style="margin-top:6px"><button id="play">Play</button>
30
+ <select id="speed" style="width:90px" aria-label="playback speed"><option value="0.5">0.5×</option><option value="1" selected>1×</option><option value="2">2×</option><option value="4">4×</option></select>
31
+ <input type="range" id="scrub" min="0" max="10" value="0" style="flex:1;min-width:120px" aria-label="time"></div>
32
+ <div id="outcome" style="margin-top:10px"></div>
33
+ <h3>Critical events (tap to jump)</h3><div id="events" class="chips"></div>
34
+ </section>
35
+ <div class="grid" style="margin-top:12px">
36
+ <section class="card"><h2>Same scene, every driver</h2><div class="scroll"><table id="cmp"></table></div></section>
37
+ <section class="card"><h2>Benchmark on unseen scenes</h2><div class="scroll"><table id="bench"></table></div><p class="small muted" id="benchnote"></p></section>
38
+ </div>
39
+ <section class="card" style="margin-top:12px">
40
+ <h2>Failure replays from the benchmark</h2>
41
+ <p class="small muted">Every recorded failure of the learned drivers can be replayed here; the simulator is deterministic.</p>
42
+ <div id="fails" class="chips"></div>
43
+ </section>
44
+ <section class="card" style="margin-top:12px">
45
+ <h2>Recorded replay</h2>
46
+ <video src="replay.mp4" controls muted playsinline poster="preview.png"></video>
47
+ </section>
48
+ <p class="status" id="status"></p>
49
+ <footer>Next Drive is a 2-D training and education simulation. It controls only speed along a fixed route, perceives the world perfectly except for line-of-sight occlusion, and has never been tested on a vehicle. It makes no claim about real-road safety.</footer>
50
+ </main>
51
+ <script src="nlrt.js"></script>
52
+ <script src="nlui.js"></script>
53
+ <script src="core.js"></script>
54
+ <script src="app.js"></script>
55
+ </body>
56
  </html>
model.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"format":"nlrt-1","meta":{"name":"Next Drive","actions":[-6.0,-3.0,-1.5,0.0,1.0,2.0],"action_names":["brake hard","brake","ease off","hold speed","speed up","accelerate"],"n_features":21,"selected_dagger_iteration":2,"note":"simulation and education only"},"nets":{"policy":[{"type":"linear","w":"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","b":"4FS3Pjs0Db3i13++Y3z6jvqkAr4mKe6+fhS2PRMojr3IjvU9RLufPjmHwrwvx2m+OLRKvCgyub3pRvo9038jvjeiAT0bQ4m+DXUxvsfYgD5zQUk+Eui9vuBCuL7d0wU+FxWTPqFtRr4RaKQ++tfUvVzuBD66ULg9UiXFvTsVMD4qCPO9GD2KvuxGzr0s2S4+CDz8vcvCtj7c77a91AhDvtU+Er2hXXU8MdxePRXpO7yl7K69+DerviKv577aEAc96YCJPQAuRD6G9+G+74A6Pt4iur7OSbW9Skg7vvkZEz4YlE4+CoGWvPvJbb353Za9X7O5PlHZQT1frbq+19EYvQ==","w_shape":[64,21],"b_shape":[64]},{"type":"relu"},{"type":"linear","w":"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","b":"3OtDPotrEz4dvuy8oYUFvjpOab6LqZCqNpoCPhaCgj5WQW29kechPltknj0Lvha9iTPXPVKWSrrCJUO8PTr4uaW7irpwOZS9wdcNvX1XHb7+pUCrzCCgPpYSVb4XD1E9wIt5vN0/gD6RvxS16b6ZsLanhj5r89akhhJkPoGfG71YYwO+LBBxPaA/DqmiQJGxN1L6PXfGN76FiP09W143tahIgD5pdcI97Qdmvn5A/D0oKte6Zi5hvRszcr5iXpA8CPydt6bPkb1GHUg96LqdPARY1LxJ6/Y7PPKePlVNx7yT9Gw+t61ZPUGKID6miUW+AqsFprqPsT7ybTs+IGBlPg==","w_shape":[64,64],"b_shape":[64]},{"type":"relu"},{"type":"linear","w":"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","b":"+VLGvE1s+74qGzS+UMSfPTfYsDtc7pY+","w_shape":[6,64],"b_shape":[6]}]},"tensors":{}}
nlrt.js ADDED
@@ -0,0 +1,219 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /* nlrt.js: browser and Node twin of nlrt.py. Batch size 1, float64 math. */
2
+ (function (root, factory) {
3
+ if (typeof module === "object" && module.exports) module.exports = factory();
4
+ else root.NLRT = factory();
5
+ })(typeof self !== "undefined" ? self : this, function () {
6
+ "use strict";
7
+
8
+ function b64d(s) {
9
+ let bin;
10
+ if (typeof atob === "function") bin = atob(s);
11
+ else bin = Buffer.from(s, "base64").toString("binary");
12
+ const bytes = new Uint8Array(bin.length);
13
+ for (let i = 0; i < bin.length; i++) bytes[i] = bin.charCodeAt(i);
14
+ return new Float32Array(bytes.buffer);
15
+ }
16
+
17
+ function decode(spec) {
18
+ for (const name in spec.nets || {}) {
19
+ for (const L of spec.nets[name]) {
20
+ if (typeof L.w === "string") L.w = b64d(L.w);
21
+ if (typeof L.b === "string") L.b = b64d(L.b);
22
+ }
23
+ }
24
+ for (const name in spec.tensors || {}) {
25
+ const t = spec.tensors[name];
26
+ if (typeof t.data === "string") t.data = b64d(t.data);
27
+ }
28
+ return spec;
29
+ }
30
+
31
+ async function loadModel(url) {
32
+ const r = await fetch(url);
33
+ if (!r.ok) throw new Error("cannot load " + url);
34
+ return decode(await r.json());
35
+ }
36
+
37
+ function pair(v) { return Array.isArray(v) ? v : [v, v]; }
38
+
39
+ // Tensor: {shape: [D]} or {shape: [C, H, W]} or {shape: [C, L]}, data: Float64Array
40
+ function T(shape, data) {
41
+ const n = shape.reduce((a, b) => a * b, 1);
42
+ return { shape: shape.slice(), data: data || new Float64Array(n) };
43
+ }
44
+
45
+ function as3(x) {
46
+ if (x.shape.length === 2) return { t: T([x.shape[0], 1, x.shape[1]], x.data), sq: true };
47
+ return { t: x, sq: false };
48
+ }
49
+
50
+ function conv2d(x0, L) {
51
+ const { t: x, sq } = as3(x0);
52
+ const [C, H, W] = x.shape;
53
+ const [O, I, KH, KW] = L.w_shape;
54
+ const [sh, sw] = pair(L.stride), [ph, pw] = pair(L.pad);
55
+ const OH = Math.floor((H + 2 * ph - KH) / sh) + 1;
56
+ const OW = Math.floor((W + 2 * pw - KW) / sw) + 1;
57
+ const out = new Float64Array(O * OH * OW);
58
+ const w = L.w, b = L.b, xd = x.data;
59
+ for (let o = 0; o < O; o++) {
60
+ for (let oy = 0; oy < OH; oy++) {
61
+ for (let ox = 0; ox < OW; ox++) {
62
+ let s = b[o];
63
+ for (let c = 0; c < I; c++) {
64
+ for (let ky = 0; ky < KH; ky++) {
65
+ const iy = oy * sh + ky - ph;
66
+ if (iy < 0 || iy >= H) continue;
67
+ const wb = ((o * I + c) * KH + ky) * KW, xb = (c * H + iy) * W;
68
+ for (let kx = 0; kx < KW; kx++) {
69
+ const ix = ox * sw + kx - pw;
70
+ if (ix < 0 || ix >= W) continue;
71
+ s += w[wb + kx] * xd[xb + ix];
72
+ }
73
+ }
74
+ }
75
+ out[(o * OH + oy) * OW + ox] = s;
76
+ }
77
+ }
78
+ }
79
+ return sq ? T([O, OW], out) : T([O, OH, OW], out);
80
+ }
81
+
82
+ function maxpool(x0, L) {
83
+ const { t: x, sq } = as3(x0);
84
+ const [C, H, W] = x.shape, [kh, kw] = pair(L.k);
85
+ const OH = Math.floor(H / kh), OW = Math.floor(W / kw);
86
+ const out = new Float64Array(C * OH * OW);
87
+ for (let c = 0; c < C; c++)
88
+ for (let oy = 0; oy < OH; oy++)
89
+ for (let ox = 0; ox < OW; ox++) {
90
+ let m = -Infinity;
91
+ for (let i = 0; i < kh; i++)
92
+ for (let j = 0; j < kw; j++) {
93
+ const v = x.data[(c * H + oy * kh + i) * W + ox * kw + j];
94
+ if (v > m) m = v;
95
+ }
96
+ out[(c * OH + oy) * OW + ox] = m;
97
+ }
98
+ return sq ? T([C, OW], out) : T([C, OH, OW], out);
99
+ }
100
+
101
+ function linear(x, L) {
102
+ const [O, I] = L.w_shape;
103
+ const out = new Float64Array(O);
104
+ for (let o = 0; o < O; o++) {
105
+ let s = L.b[o];
106
+ const base = o * I;
107
+ for (let i = 0; i < I; i++) s += L.w[base + i] * x.data[i];
108
+ out[o] = s;
109
+ }
110
+ return T([O], out);
111
+ }
112
+
113
+ function map(x, f) {
114
+ const d = new Float64Array(x.data.length);
115
+ for (let i = 0; i < d.length; i++) d[i] = f(x.data[i]);
116
+ return T(x.shape, d);
117
+ }
118
+
119
+ function forward(layers, x) {
120
+ if (!(x.data instanceof Float64Array)) x = T(x.shape, Float64Array.from(x.data));
121
+ for (const L of layers) {
122
+ switch (L.type) {
123
+ case "linear": x = linear(x, L); break;
124
+ case "conv2d": x = conv2d(x, L); break;
125
+ case "relu": x = map(x, v => (v > 0 ? v : 0)); break;
126
+ case "tanh": x = map(x, Math.tanh); break;
127
+ case "sigmoid": x = map(x, v => 1 / (1 + Math.exp(-v))); break;
128
+ case "maxpool": x = maxpool(x, L); break;
129
+ case "gap": {
130
+ const C = x.shape[0], n = x.data.length / C, d = new Float64Array(C);
131
+ for (let c = 0; c < C; c++) { let s = 0; for (let i = 0; i < n; i++) s += x.data[c * n + i]; d[c] = s / n; }
132
+ x = T([C], d); break;
133
+ }
134
+ case "flatten": x = T([x.data.length], x.data); break;
135
+ case "reshape": x = T(L.shape, x.data); break;
136
+ case "upsample": {
137
+ const s = L.s;
138
+ if (x.shape.length === 3) {
139
+ const [C, H, W] = x.shape, d = new Float64Array(C * H * s * W * s);
140
+ for (let c = 0; c < C; c++)
141
+ for (let y = 0; y < H * s; y++)
142
+ for (let xx = 0; xx < W * s; xx++)
143
+ d[(c * H * s + y) * W * s + xx] = x.data[(c * H + Math.floor(y / s)) * W + Math.floor(xx / s)];
144
+ x = T([C, H * s, W * s], d);
145
+ } else {
146
+ const [C, Ln] = x.shape, d = new Float64Array(C * Ln * s);
147
+ for (let c = 0; c < C; c++) for (let i = 0; i < Ln * s; i++) d[c * Ln * s + i] = x.data[c * Ln + Math.floor(i / s)];
148
+ x = T([C, Ln * s], d);
149
+ }
150
+ break;
151
+ }
152
+ case "l2norm": {
153
+ let s = 0; for (const v of x.data) s += v * v;
154
+ const n = Math.max(Math.sqrt(s), 1e-12);
155
+ x = map(x, v => v / n); break;
156
+ }
157
+ default: throw new Error("unknown layer " + L.type);
158
+ }
159
+ }
160
+ return x;
161
+ }
162
+
163
+ function mulberry32(seed) {
164
+ let a = seed >>> 0;
165
+ const random = function () {
166
+ a = (a + 0x6D2B79F5) >>> 0;
167
+ let t = a;
168
+ t = Math.imul(t ^ (t >>> 15), t | 1) >>> 0;
169
+ t = (t ^ ((t + (Math.imul(t ^ (t >>> 7), t | 61) >>> 0)) >>> 0)) >>> 0;
170
+ return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
171
+ };
172
+ return {
173
+ random,
174
+ uniform: (lo, hi) => lo + (hi - lo) * random(),
175
+ randint: n => Math.floor(random() * n),
176
+ normal: () => { const u = Math.max(random(), 1e-12), v = random(); return Math.sqrt(-2 * Math.log(u)) * Math.cos(2 * Math.PI * v); },
177
+ };
178
+ }
179
+
180
+ function softmax(arr) {
181
+ let m = -Infinity; for (const v of arr) if (v > m) m = v;
182
+ const e = Array.from(arr, v => Math.exp(v - m)); const s = e.reduce((a, b) => a + b, 0);
183
+ return e.map(v => v / s);
184
+ }
185
+
186
+ function argmax(arr) { let k = 0; for (let i = 1; i < arr.length; i++) if (arr[i] > arr[k]) k = i; return k; }
187
+
188
+ // Deep comparison used by parity tests: numbers within tol, everything else exact.
189
+ function compare(got, want, tol, path, acc) {
190
+ acc = acc || { maxDiff: 0, mismatches: [] };
191
+ path = path || "$";
192
+ if (typeof want === "number") {
193
+ const d = Math.abs(Number(got) - want);
194
+ if (!(d <= tol * Math.max(1, Math.abs(want)))) acc.mismatches.push(path + ": got " + got + " want " + want);
195
+ if (isFinite(d)) acc.maxDiff = Math.max(acc.maxDiff, d);
196
+ } else if (Array.isArray(want)) {
197
+ if (!got || got.length !== want.length) acc.mismatches.push(path + ": length " + (got && got.length) + " vs " + want.length);
198
+ else for (let i = 0; i < want.length; i++) compare(got[i], want[i], tol, path + "[" + i + "]", acc);
199
+ } else if (want && typeof want === "object") {
200
+ for (const k in want) compare(got ? got[k] : undefined, want[k], tol, path + "." + k, acc);
201
+ } else if (got !== want) acc.mismatches.push(path + ": got " + got + " want " + want);
202
+ return acc;
203
+ }
204
+
205
+ function runParity(core, ctx, parity) {
206
+ const results = [];
207
+ let maxDiff = 0, failed = 0;
208
+ for (const c of parity.cases) {
209
+ const out = core.runCase(ctx, c.input);
210
+ const r = compare(out, c.expected, c.tol != null ? c.tol : parity.tol);
211
+ maxDiff = Math.max(maxDiff, r.maxDiff);
212
+ if (r.mismatches.length) failed++;
213
+ results.push({ name: c.name, maxDiff: r.maxDiff, mismatches: r.mismatches.slice(0, 5) });
214
+ }
215
+ return { cases: parity.cases.length, failed, maxDiff, results };
216
+ }
217
+
218
+ return { b64d, decode, loadModel, T, forward, mulberry32, softmax, argmax, compare, runParity };
219
+ });
nlui.js ADDED
@@ -0,0 +1,123 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ /* nlui.js: small DOM and canvas helpers shared by Next Lab Spaces. */
2
+ (function (root) {
3
+ "use strict";
4
+ const $ = id => document.getElementById(id);
5
+ function h(tag, attrs, ...kids) {
6
+ const e = document.createElement(tag);
7
+ for (const k in attrs || {}) {
8
+ if (k === "class") e.className = attrs[k];
9
+ else if (k.startsWith("on")) e.addEventListener(k.slice(2), attrs[k]);
10
+ else if (k === "html") e.innerHTML = attrs[k];
11
+ else e.setAttribute(k, attrs[k]);
12
+ }
13
+ for (const c of kids.flat()) if (c != null) e.append(c.nodeType ? c : document.createTextNode(String(c)));
14
+ return e;
15
+ }
16
+ const css = name => getComputedStyle(document.documentElement).getPropertyValue(name).trim();
17
+ const fmt = (v, d = 3) => (v == null || !isFinite(v) ? "–" : Number(v).toFixed(d));
18
+
19
+ function grayImage(canvas, data, w, h, scale, opts) {
20
+ opts = opts || {};
21
+ canvas.width = w * scale; canvas.height = h * scale;
22
+ const ctx = canvas.getContext("2d");
23
+ const img = ctx.createImageData(w, h);
24
+ let lo = opts.lo != null ? opts.lo : 0, hi = opts.hi != null ? opts.hi : 1;
25
+ for (let i = 0; i < w * h; i++) {
26
+ const v = Math.max(0, Math.min(255, Math.round(255 * (data[i] - lo) / (hi - lo || 1))));
27
+ img.data[4 * i] = img.data[4 * i + 1] = img.data[4 * i + 2] = v; img.data[4 * i + 3] = 255;
28
+ }
29
+ const tmp = document.createElement("canvas"); tmp.width = w; tmp.height = h;
30
+ tmp.getContext("2d").putImageData(img, 0, 0);
31
+ ctx.imageSmoothingEnabled = false;
32
+ ctx.drawImage(tmp, 0, 0, w * scale, h * scale);
33
+ return ctx;
34
+ }
35
+
36
+ function rgbImage(canvas, data, w, h, scale) { // data: CHW in [0,1]
37
+ canvas.width = w * scale; canvas.height = h * scale;
38
+ const ctx = canvas.getContext("2d");
39
+ const img = ctx.createImageData(w, h), n = w * h;
40
+ for (let i = 0; i < n; i++) {
41
+ for (let c = 0; c < 3; c++) img.data[4 * i + c] = Math.max(0, Math.min(255, Math.round(255 * data[c * n + i])));
42
+ img.data[4 * i + 3] = 255;
43
+ }
44
+ const tmp = document.createElement("canvas"); tmp.width = w; tmp.height = h;
45
+ tmp.getContext("2d").putImageData(img, 0, 0);
46
+ ctx.imageSmoothingEnabled = false;
47
+ ctx.drawImage(tmp, 0, 0, w * scale, h * scale);
48
+ return ctx;
49
+ }
50
+
51
+ // series: [{name, color, points: [[x,y],...], dash}]
52
+ function lineChart(canvas, series, opts) {
53
+ opts = opts || {};
54
+ const W = canvas.clientWidth || 600, H = opts.height || 240, dpr = window.devicePixelRatio || 1;
55
+ canvas.width = W * dpr; canvas.height = H * dpr; canvas.style.height = H + "px";
56
+ const ctx = canvas.getContext("2d"); ctx.scale(dpr, dpr);
57
+ const m = { l: 46, r: 10, t: 10, b: 34 };
58
+ let xs = [], ys = [];
59
+ for (const s of series) for (const [x, y] of s.points) { if (isFinite(y)) { xs.push(x); ys.push(y); } }
60
+ if (opts.hlines) for (const hl of opts.hlines) ys.push(hl.y);
61
+ let x0 = opts.xmin != null ? opts.xmin : Math.min(...xs), x1 = opts.xmax != null ? opts.xmax : Math.max(...xs);
62
+ let y0 = opts.ymin != null ? opts.ymin : Math.min(...ys), y1 = opts.ymax != null ? opts.ymax : Math.max(...ys);
63
+ if (y1 - y0 < 1e-9) { y0 -= 1; y1 += 1; }
64
+ if (x1 - x0 < 1e-9) x1 = x0 + 1;
65
+ const X = x => m.l + (x - x0) / (x1 - x0) * (W - m.l - m.r);
66
+ const Y = y => H - m.b - (y - y0) / (y1 - y0) * (H - m.t - m.b);
67
+ ctx.font = "11px system-ui"; ctx.fillStyle = css("--muted"); ctx.strokeStyle = css("--line"); ctx.lineWidth = 1;
68
+ for (let i = 0; i <= 4; i++) {
69
+ const y = y0 + (y1 - y0) * i / 4;
70
+ ctx.beginPath(); ctx.moveTo(m.l, Y(y)); ctx.lineTo(W - m.r, Y(y)); ctx.stroke();
71
+ ctx.fillText(Math.abs(y) >= 100 ? y.toFixed(0) : y.toFixed(2), 4, Y(y) + 4);
72
+ }
73
+ for (let i = 0; i <= 4; i++) {
74
+ const x = x0 + (x1 - x0) * i / 4;
75
+ ctx.fillText(Math.abs(x) >= 100 || Number.isInteger(x) ? x.toFixed(0) : x.toFixed(1), X(x) - 8, H - m.b + 14);
76
+ }
77
+ if (opts.xlabel) ctx.fillText(opts.xlabel, W / 2 - 30, H - 4);
78
+ if (opts.hlines) for (const hl of opts.hlines) {
79
+ ctx.setLineDash([4, 4]); ctx.strokeStyle = hl.color || css("--muted");
80
+ ctx.beginPath(); ctx.moveTo(m.l, Y(hl.y)); ctx.lineTo(W - m.r, Y(hl.y)); ctx.stroke(); ctx.setLineDash([]);
81
+ }
82
+ for (const s of series) {
83
+ ctx.strokeStyle = s.color; ctx.lineWidth = s.width || 2; ctx.setLineDash(s.dash || []);
84
+ ctx.beginPath(); let first = true;
85
+ for (const [x, y] of s.points) { if (!isFinite(y)) continue; if (first) ctx.moveTo(X(x), Y(y)); else ctx.lineTo(X(x), Y(y)); first = false; }
86
+ ctx.stroke(); ctx.setLineDash([]);
87
+ }
88
+ return { X, Y, ctx };
89
+ }
90
+
91
+ function legend(el, series) {
92
+ el.innerHTML = "";
93
+ for (const s of series) el.append(h("span", { class: "small", style: "margin-right:12px;white-space:nowrap" },
94
+ h("span", { style: `display:inline-block;width:14px;height:3px;background:${s.color};vertical-align:middle;margin-right:4px` }), s.name));
95
+ }
96
+
97
+ async function loadAssets(core) {
98
+ const assets = {};
99
+ for (const name of core.ASSETS) {
100
+ const r = await fetch(name);
101
+ if (!r.ok) throw new Error("cannot load " + name);
102
+ assets[name] = await r.json();
103
+ }
104
+ return assets;
105
+ }
106
+
107
+ // Standard bootstrap: load assets, init core, expose __parity for automated tests.
108
+ async function boot(core, start) {
109
+ try {
110
+ const assets = await loadAssets(core);
111
+ const ctx = core.init(assets);
112
+ const parity = await (await fetch("parity_cases.json")).json();
113
+ root.__parity = () => NLRT.runParity(core, ctx, parity);
114
+ await start(ctx, assets);
115
+ root.__ready = true;
116
+ } catch (e) {
117
+ console.error(e); root.__error = String(e && e.stack || e);
118
+ const s = document.getElementById("status"); if (s) s.textContent = "Failed to start: " + e;
119
+ }
120
+ }
121
+
122
+ root.NLUI = { $, h, css, fmt, grayImage, rgbImage, lineChart, legend, boot };
123
+ })(typeof self !== "undefined" ? self : this);
parity_cases.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"tol": 0, "cases": [{"name": "10007-mixed-2-shield", "input": {"seed": 10007, "kind": "mixed", "density": 2, "driver": "shield"}, "expected": {"route": "straight", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 3, "overrides": 39, "steps": 344, "actions_hash": 448908544}}, {"name": "10007-mixed-2-learned", "input": {"seed": 10007, "kind": "mixed", "density": 2, "driver": "learned"}, "expected": {"route": "straight", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 0, "overrides": 0, "steps": 189, "actions_hash": 668221100}}, {"name": "12345-occlusion-3-shield", "input": {"seed": 12345, "kind": "occlusion", "density": 3, "driver": "shield"}, "expected": {"route": "straight", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 0, "overrides": 0, "steps": 312, "actions_hash": 156160153}}, {"name": "10111-lead_brake-1-learned", "input": {"seed": 10111, "kind": "lead_brake", "density": 1, "driver": "learned"}, "expected": {"route": "straight", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 2, "overrides": 0, "steps": 149, "actions_hash": 297292238}}, {"name": "10222-crosswalk-2-rule", "input": {"seed": 10222, "kind": "crosswalk", "density": 2, "driver": "rule"}, "expected": {"route": "left", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 4, "overrides": 0, "steps": 398, "actions_hash": 292331991}}, {"name": "10333-light-3-expert", "input": {"seed": 10333, "kind": "light", "density": 3, "driver": "expert"}, "expected": {"route": "right", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 7, "overrides": 0, "steps": 244, "actions_hash": 557514203}}, {"name": "10444-occlusion-2-rule", "input": {"seed": 10444, "kind": "occlusion", "density": 2, "driver": "rule"}, "expected": {"route": "straight", "collision": true, "red_violations": 0, "reached": false, "hard_brakes": 1, "overrides": 0, "steps": 41, "actions_hash": 191279273}}, {"name": "10555-mixed-3-shield", "input": {"seed": 10555, "kind": "mixed", "density": 3, "driver": "shield"}, "expected": {"route": "straight", "collision": false, "red_violations": 0, "reached": true, "hard_brakes": 0, "overrides": 0, "steps": 234, "actions_hash": 582273325}}]}
preview.png ADDED
replay.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2a1ac18bbb78c61d07c6013e9ce42285c0aefc9181d5fb3c7211468e9d30fa09
3
+ size 153163
style.css CHANGED
@@ -1,28 +1,55 @@
1
- body {
2
- padding: 2rem;
3
- font-family: -apple-system, BlinkMacSystemFont, "Arial", sans-serif;
 
4
  }
5
-
6
- h1 {
7
- font-size: 16px;
8
- margin-top: 0;
9
- }
10
-
11
- p {
12
- color: rgb(107, 114, 128);
13
- font-size: 15px;
14
- margin-bottom: 10px;
15
- margin-top: 5px;
16
- }
17
-
18
- .card {
19
- max-width: 620px;
20
- margin: 0 auto;
21
- padding: 16px;
22
- border: 1px solid lightgray;
23
- border-radius: 16px;
24
- }
25
-
26
- .card p:last-child {
27
- margin-bottom: 0;
28
  }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ :root {
2
+ --bg: #f7f7f5; --panel: #ffffff; --ink: #1d1f22; --muted: #5f6670; --line: #dcdfe3;
3
+ --accent: #0b6e4f; --accent-soft: #e3f1ec; --warn: #a4471b; --warn-soft: #fbeee6;
4
+ --blue: #2f5fb3; --gold: #b07d12; --red: #c0392b;
5
  }
6
+ @media (prefers-color-scheme: dark) {
7
+ :root {
8
+ --bg: #15171a; --panel: #1e2125; --ink: #eceef0; --muted: #a5acb5; --line: #33383e;
9
+ --accent: #4cc39a; --accent-soft: #1b3229; --warn: #f0a070; --warn-soft: #3a2519;
10
+ --blue: #7ea6f0; --gold: #e0b44c; --red: #f07a6c;
11
+ }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
12
  }
13
+ * { box-sizing: border-box; }
14
+ html, body { margin: 0; padding: 0; background: var(--bg); color: var(--ink); }
15
+ body { font: 15px/1.5 system-ui, -apple-system, "Segoe UI", Roboto, sans-serif; }
16
+ main { max-width: 1040px; margin: 0 auto; padding: 16px; }
17
+ header h1 { font-size: 1.6rem; margin: 4px 0 2px; letter-spacing: -0.01em; }
18
+ header p { margin: 0 0 8px; color: var(--muted); max-width: 70ch; }
19
+ .badges { display: flex; flex-wrap: wrap; gap: 6px; margin: 6px 0 12px; }
20
+ .badge { font-size: 12px; padding: 2px 8px; border-radius: 999px; background: var(--accent-soft); color: var(--accent); border: 1px solid var(--line); }
21
+ .badge.warn { background: var(--warn-soft); color: var(--warn); }
22
+ .grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(300px, 1fr)); gap: 12px; }
23
+ .card { background: var(--panel); border: 1px solid var(--line); border-radius: 10px; padding: 12px 14px; min-width: 0; }
24
+ .card h2 { font-size: 1.02rem; margin: 0 0 8px; }
25
+ .card h3 { font-size: .95rem; margin: 10px 0 6px; }
26
+ label { display: block; font-size: 13px; color: var(--muted); margin: 8px 0 3px; }
27
+ input[type=text], textarea, select { width: 100%; font: inherit; padding: 7px 9px; border-radius: 8px; border: 1px solid var(--line); background: var(--bg); color: var(--ink); }
28
+ textarea { min-height: 110px; resize: vertical; }
29
+ input[type=range] { width: 100%; }
30
+ button { font: inherit; padding: 7px 12px; border-radius: 8px; border: 1px solid var(--accent); background: var(--accent); color: #fff; cursor: pointer; }
31
+ button.ghost { background: transparent; color: var(--accent); }
32
+ button:disabled { opacity: .5; cursor: default; }
33
+ .row { display: flex; flex-wrap: wrap; gap: 8px; align-items: center; }
34
+ canvas { max-width: 100%; height: auto; display: block; border-radius: 6px; }
35
+ .card > canvas { width: 100%; }
36
+ .muted { color: var(--muted); }
37
+ .small { font-size: 13px; }
38
+ .status { font-size: 13px; color: var(--muted); min-height: 1.4em; }
39
+ table { border-collapse: collapse; width: 100%; font-size: 13px; }
40
+ th, td { text-align: left; padding: 4px 6px; border-bottom: 1px solid var(--line); vertical-align: top; }
41
+ th { color: var(--muted); font-weight: 600; }
42
+ td.num, th.num { text-align: right; font-variant-numeric: tabular-nums; }
43
+ .scroll { overflow-x: auto; }
44
+ mark { background: #ffe58a; color: #1d1f22; padding: 0 2px; border-radius: 3px; }
45
+ mark.weak { background: #fff3c4; }
46
+ video { width: 100%; border-radius: 8px; }
47
+ footer { margin: 18px 0 8px; font-size: 13px; color: var(--muted); }
48
+ a { color: var(--blue); }
49
+ .chips { display: flex; flex-wrap: wrap; gap: 6px; }
50
+ .chip { border: 1px solid var(--line); border-radius: 999px; padding: 3px 10px; background: var(--bg); font-size: 13px; cursor: pointer; color: var(--ink); }
51
+ .chip.on { border-color: var(--accent); background: var(--accent-soft); }
52
+ .kpi { display: grid; grid-template-columns: repeat(auto-fit, minmax(120px, 1fr)); gap: 8px; }
53
+ .kpi div { background: var(--bg); border: 1px solid var(--line); border-radius: 8px; padding: 6px 8px; }
54
+ .kpi b { display: block; font-size: 1.15rem; }
55
+ .kpi span { font-size: 12px; color: var(--muted); }