feat(simulation): add Worker AI polygon runtime and terrain navigation

2026-09-25 16:40:45 +03:00
parent a7c64e009d
commit f01bd39037
88 changed files with 9918 additions and 108 deletions
@@ -0,0 +1,57 @@
+"""Offline numeric check of the pinned GOOSE decoder on retained Worker RGB.
+
+Runs in the existing pinned image, without changing its runner or checkpoint.
+This diagnoses the adapter; it is not semantic accuracy or navigation acceptance.
+"""
+
+import argparse
+import csv
+import hashlib
+import importlib.util
+import json
+import time
+from pathlib import Path
+
+import numpy as np
+import torch
+from PIL import Image
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--image", type=Path, required=True)
+    parser.add_argument("--output", type=Path, required=True)
+    args = parser.parse_args()
+    spec = importlib.util.spec_from_file_location("reference", "/assets/ddrnet-goose-runner.py")
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    model, _, _ = module.load_model("ddrnet", Path("/assets/ddrnet-checkpoint.pth"))
+    tensor, _ = module.preprocess(Image.open(args.image))
+    tensor = tensor.cuda()
+    with torch.inference_mode():
+        logits = module.logits_from_output(model(tensor)).float()
+        legacy = torch.sigmoid(logits).argmax(1)
+        direct = logits.argmax(1)
+        saturated = (torch.sigmoid(logits) == 1).sum(1)
+    names = {}
+    with open("/assets/ddrnet-goose-mapping.csv") as stream:
+        names = {int(r["label_key"]): r["class_name"] for r in csv.DictReader(stream)}
+    args.output.mkdir(exist_ok=True, parents=True)
+    report = {
+        "source_sha256": hashlib.sha256(args.image.read_bytes()).hexdigest(),
+        "monotonic_ns": time.monotonic_ns(),
+        "logit_range": [float(logits.min()), float(logits.max())],
+        "changed_pixels": int((legacy != direct).sum()),
+        "saturated_tie_pixels": int((saturated > 1).sum()),
+    }
+    for name, mask in (("reference", legacy), ("direct", direct)):
+        mask = mask[0].cpu().numpy().astype(np.uint8)
+        Image.fromarray(mask).save(args.output / (name + ".png"))
+        ids, counts = np.unique(mask, return_counts=True)
+        report[name] = {names[int(i)]: int(counts[index]) for index, i in enumerate(ids)}
+    (args.output / "report.json").write_text(json.dumps(report, indent=2), encoding="utf-8")
+    print(json.dumps(report))
+
+
+if __name__ == "__main__":
+    main()