-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
1302 lines (1171 loc) · 61.8 KB
/
index.html
File metadata and controls
1302 lines (1171 loc) · 61.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!doctype html>
<html lang="en">
<head>
<style>
pre, code {
font-size: 16px;
}
</style>
<style>
.leaderboard-table thead th:nth-child(3),
.leaderboard-table thead th:nth-child(4),
.leaderboard-table thead th:nth-child(12),
.leaderboard-table thead th:nth-child(13),
.leaderboard-table tbody td:nth-child(3),
.leaderboard-table tbody td:nth-child(4),
.leaderboard-table tbody td:nth-child(12),
.leaderboard-table tbody td:nth-child(13) {
display: none;
}
.leaderboard-table { border-radius: 10px; overflow: hidden; border: 0; }
.leaderboard-table thead { background: linear-gradient(90deg, #ff6b6b, #f7b267); color: #fff; }
.leaderboard-table th { text-transform: uppercase; letter-spacing: .04em; font-weight: 700; }
.leaderboard-table tbody tr:nth-child(odd) { background: #fffaf0; }
.leaderboard-table tbody tr:nth-child(even) { background: #fff5f7; }
.leaderboard-table tbody tr:hover { background: #e0f2fe; }
.section-header { background-color: #b2f5ea; color: #234e52; }
.human-level-row { background: #fefcbf !important; }
.best-score { color: #d53f8c; text-decoration-color: #ed64a6; }
.model-badge.proprietary-badge { background: #fed7e2; color: #702459; border-color: #fbb6ce; }
.model-badge.open-badge { background: #c6f6d5; color: #22543d; border-color: #9ae6b4; }
.model-badge.light-nav { background: #fef3c7; color: #92400e; border-color: #fde68a; }
.model-badge.navllm { background: #e0e7ff; color: #3730a3; border-color: #c7d2fe; }
</style>
<title>NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions</title>
<link rel="icon" type="image/x-icon" href="./static/img/icons/brain.ico">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<!-- Open Graph -->
<meta property="og:url" content="https://vision-x-nyu.github.io/thinking-in-space.github.io/" />
<meta property="og:image" content="./static/img/preview.png" />
<meta property="og:title" content="NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions" />
<!-- Twitter -->
<meta name="twitter:url" content="https://vision-x-nyu.github.io/thinking-in-space.github.io/" />
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:image" content="./static/img/preview.png" />
<meta name="twitter:title" content="NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions" />
<meta name="twitter:description" content="We introduce NavSpace, a novel benchmark of over 1200 video-based visual-spatial intelligence questions, to evaluate and probe MLLMs, which revealed that their emerging spatial reasoning and local world modeling capabilities remain subhuman but promising." />
<!-- JS Libraries -->
<script src="./static/js/distill_template.v2.js"></script>
<script src="https://polyfill.io/v3/polyfill.min.js?features=es6"></script>
<script id="MathJax-script" async src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js"></script>
<script src="https://d3js.org/d3.v5.min.js"></script>
<script src="https://d3js.org/d3-collection.v1.min.js"></script>
<script src="https://rawgit.com/nstrayer/slid3r/master/dist/slid3r.js"></script>
<script defer="" src="./static/js/hider.js"></script>
<script defer src="./static/js/image_interact.js"></script>
<script defer src="./static/js/switch_videos.js"></script>
<script defer src="./static/js/video-speed.js"></script>
<!-- Stylesheets -->
<link rel="stylesheet" href="./static/css/style.css">
<link rel="stylesheet" href="./static/css/fontawesome.all.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<!-- KaTeX -->
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/katex.min.css" integrity="sha384-yFRtMMDnQtDRO8rLpMIKrtPCD5jdktao2TV19YiZYWMDkUR5GQZR/NOVTdquEx1j" crossorigin="anonymous">
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/katex.min.js" integrity="sha384-9Nhn55MVVN0/4OFx7EE5kpFBPsEMZxKTCnA+4fqDmg12eCTqGi6+BB2LjY8brQxJ" crossorigin="anonymous"></script>
<script defer src="https://cdn.jsdelivr.net/npm/katex@0.10.2/dist/contrib/auto-render.min.js" integrity="sha384-kWPLUVMOks5AQFrykwIup5lo0m3iMkkHrD0uJ4H5cjeGihAutqP0yW0J6dpFiVkI" crossorigin="anonymous" onload="renderMathInElement(document.body);"></script>
<!-- FontAwesome -->
<script defer src="./static/js/fontawesome.all.min.js"></script>
<!-- Medium Zoom -->
<script src="https://cdn.jsdelivr.net/npm/jquery@3.7.1/dist/jquery.min.js"></script>
<script defer src="./static/js/medium-zoom.min.js"></script>
<script defer src="./static/js/zoom.js"></script>
<!-- Leaderboard Styles -->
<style>
.leaderboard-table {
width: 140%;
margin-left: -20%;
border-collapse: separate;
border-spacing: 0;
background: #ffffff;
border-radius: 8px;
overflow: hidden;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, sans-serif;
font-size: 14px;
margin-bottom: 40px;
}
.leaderboard-table th:first-child,
.leaderboard-table td:first-child {
width: 25%;
min-width: 200px;
}
.leaderboard-container {
position: relative;
}
.leaderboard-table thead {
background: #4a5568;
color: white;
}
.leaderboard-table th {
padding: 8px 6px;
text-align: center;
font-weight: 600;
font-size: 12px;
letter-spacing: 0.025em;
border-bottom: 1px solid rgba(255, 255, 255, 0.1);
position: relative;
}
.sortable-header {
cursor: pointer;
user-select: none;
}
.sortable-header .sort-indicator {
position: absolute;
right: 8px;
top: 50%;
font-size: 14px;
line-height: 1;
transform: translateY(-50%);
color: #a0aec0;
}
.sortable-header.active .sort-indicator {
color: white;
}
.section-header {
background-color: #deeef8;
font-weight: 700;
color: #2d3748;
letter-spacing: 1px;
font-size: 18px;
text-align: center;
padding: 12px !important;
border-bottom: 2px solid #e2e8f0;
}
.leaderboard-table td {
padding: 6px 4px;
border-bottom: 1px solid #e2e8f0;
color: #4a5568;
text-align: center;
font-size: 13px;
}
.leaderboard-table tbody tr {
transition: background-color 0.15s ease;
}
.leaderboard-table tbody tr:hover {
background-color: #f7fafc;
}
.leaderboard-table td:first-child {
font-weight: 600;
text-align: left;
color: #2d3748;
padding-left: 8px;
}
.leaderboard-table td:nth-child(5) {
font-weight: 700;
font-size: 15px;
color: #2d3748;
}
.human-level-row {
background: #fffbeb85 !important;
font-weight: 600;
}
.top-performer {
background: #f0fdf494 !important;
}
.top-performer td:first-child::before {
content: '🥇 ';
}
.model-badge {
display: inline-block;
padding: 2px 6px;
border-radius: 4px;
font-size: 11px;
font-weight: 500;
margin-left: 8px;
text-transform: uppercase;
letter-spacing: 0.025em;
}
.proprietary-badge {
background: #e2e8f0;
color: #475569;
border: 1px solid #cbd5e0;
}
.open-badge {
background: #e6fffa;
color: #047857;
border: 1px solid #a7f3d0;
}
.leaderboard-table tbody tr:nth-child(even) {
background-color: #fafafa;
}
.baseline-row {
font-style: italic;
opacity: 0.8;
background-color: #f9fafb;
}
.best-score {
font-weight: 700;
text-decoration: underline;
text-decoration-color: #cbd5e0;
text-underline-offset: 2px;
}
.section-divider {
border-top: 2px solid #e2e8f0;
}
@media (max-width: 768px) {
.leaderboard-container {
overflow-x: auto;
-webkit-overflow-scrolling: touch;
scroll-behavior: smooth;
border-radius: 8px;
box-shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
}
.leaderboard-table {
width: 1200px; /* Fixed width to ensure horizontal scrolling */
margin-left: 0;
font-size: 12px;
min-width: 1200px;
}
.leaderboard-table th,
.leaderboard-table td {
padding: 8px;
white-space: nowrap;
min-width: 80px;
}
.leaderboard-table td:first-child {
min-width: 250px;
max-width: 250px;
overflow: hidden;
text-overflow: ellipsis;
padding-left: 12px;
}
/* Add scroll hint for mobile users */
.leaderboard-container::after {
content: "← Swipe to scroll →";
position: absolute;
bottom: -25px;
left: 50%;
transform: translateX(-50%);
font-size: 11px;
color: #666;
font-style: italic;
pointer-events: none;
}
}
</style>
<script>
// Fallbacks for missing assets so the page can render without local files
(function () {
const PLACEHOLDER_SRC = './static/img/placeholder.svg';
function applyImgFallback(img) {
if (!img || img.dataset.fallbackApplied === '1') return;
img.dataset.fallbackApplied = '1';
img.removeAttribute('srcset');
img.src = PLACEHOLDER_SRC;
}
document.addEventListener('error', function (e) {
const target = e.target;
if (target && target.tagName === 'IMG') {
applyImgFallback(target);
}
}, true);
document.addEventListener('DOMContentLoaded', function () {
// Ensure all existing <img> have fallbacks
document.querySelectorAll('img').forEach(function (img) {
img.addEventListener('error', function () { applyImgFallback(img); });
if (img.complete && img.naturalWidth === 0) {
applyImgFallback(img);
}
});
// Video fallback: replace broken videos with a placeholder image
function replaceVideoWithPlaceholder(videoEl) {
if (!videoEl || videoEl.dataset.fallbackApplied === '1') return;
videoEl.dataset.fallbackApplied = '1';
const img = document.createElement('img');
img.alt = 'placeholder';
img.src = PLACEHOLDER_SRC;
img.style.width = '100%';
img.style.height = 'auto';
try {
videoEl.parentNode.replaceChild(img, videoEl);
} catch (_) { /* no-op */ }
}
document.querySelectorAll('video').forEach(function (vid) {
// If source fails to load, swap with placeholder
const source = vid.querySelector('source');
if (source) {
source.addEventListener('error', function () { replaceVideoWithPlaceholder(vid); }, true);
}
vid.addEventListener('error', function () { replaceVideoWithPlaceholder(vid); }, true);
// Give the browser a moment; if still not loaded, fallback
setTimeout(function () {
if (vid.readyState === 0) {
replaceVideoWithPlaceholder(vid);
}
}, 1200);
});
// Define no-op handlers to avoid errors if supporting scripts are missing
if (typeof window.switchVideo !== 'function') {
window.switchVideo = function () { /* no-op for missing script */ };
}
});
})();
</script>
</head>
<body>
<div class="header-wrapper">
<div class="header-container" id="header-container">
<div class="header-content">
<h1 style="margin-top: 0px"><i>NavSpace</i></h1>
<div class="responsive-header">
<h2>How Navigation Agents Follow Spatial Intelligence Instructions</h2>
</div>
<div class="icon-container">
<div class="icon-item">
<img src="static/img/icons/benchmark.svg" alt="Visual Representation Icon">
<div><strong>NavSpace</strong>: We introduce the first spatial intelligence benchmark for instruction-based navigation: NavSpace.</div>
</div>
<div class="icon-item">
<img src="./static/img/icons/evaluation.svg" alt="Connector Design Icon">
<div><strong>Evaluation</strong>: we comprehensively evaluate 22 navigation agents in total, containing models from open-source MLLM, proprietary MLLM, lightweight navigation models to navigation large models.</div>
</div>
<div class="icon-item">
<img src="./static/img/icons/speech.svg" alt="Speech Logo" class="icon">
<div><strong>Discussions</strong>: Based on the evaluation results, we conducted a detailed analysis of the limitations of existing methods and distilled four key insights.</div>
</div>
<div class="icon-item">
<img src="static/img/icons/baseline.svg" alt="Vision Logo" class="icon">
<div><strong>Baseline Model: SNav</strong>: We propose SNav, a spatially intelligent navigation model, that surpasses existing models and establishes a strong baseline for NavSpace and real robot tests. </div>
</div>
</div>
<div class="button-container">
<a href="https://arxiv.org/abs/2510.08173" class="button paper-link" target="_blank">
<span class="icon is-small">
<img src="static/img/icons/arxiv.svg" alt="Github_logo" style="height: 1.5em;">
</span>
arXiv
</a>
<a href="https://github.com/TidalHarley/NavSpace-How-Navigation-Agents-Follow-Spatial-Intelligence-Instructions" class="button" target="_blank">
<span class="icon is-small">
<img src="static/img/icons/github.svg" alt="Github_logo" style="height: 1.5em;">
</span>
<span>Code</span>
</a>
<a href="https://huggingface.co" class="button" target="_blank">
<span class="icon is-small">
<img src="static/img/icons/huggingface.svg" alt="Hugging Face logo" style="height: 1.5em;">
</span>
<span>SNav Model(Coming Soon)</span>
</a>
<a href="#NavSpace-Leaderboard" class="button">
<span class="icon is-small">
<img src="static/img/icons/Leaderboard.svg" alt="Leaderboard logo" style="height: 1.5em;">
</span>
<span>Leaderboard</span>
</a>
</div>
</div>
<div class="header-image">
<img draggable="false" src="static/img/preview.png" alt="Teaser Image" class="teaser-image">
</div>
</div>
</div>
<d-article>
<div class="byline">
<div style="max-width: 1000px; margin: 0 auto; padding: 20px;">
<!-- Authors Section -->
<div style="text-align: center; margin-bottom: 20px;">
<h3 style="font-size: 1.2em; margin-bottom: 15px;"></h3>
<div style="line-height: 1.6; font-size: 0.85em; white-space: nowrap; overflow: hidden;">
<div style="margin-bottom: 5px;">
<a class="author-link" target="_blank">Haolin Yang</a><sup>1,2*</sup>,
<a class="author-link" target="_blank">Yuxing Long</a><sup>1,2*</sup>,
<a class="author-link" target="_blank">Zhuoyuan Yu</a><sup>1,2</sup>,
<a class="author-link" target="_blank">Zihan Yang</a><sup>1,2</sup>,
<a class="author-link" target="_blank">Minghan Wang</a><sup>1</sup>
</div>
<div>
<a class="author-link" target="_blank">Jiapeng Xu</a><sup>1</sup>,
<a class="author-link" target="_blank">Yihan Wang</a><sup>1</sup>,
<a class="author-link" target="_blank">Ziyan Yu</a><sup>1</sup>,
<a class="author-link" target="_blank">Wenzhe Cai</a><sup>3</sup>,
<a class="author-link" target="_blank">Lei Kang</a><sup>2</sup>,
<a class="author-link" target="_blank">Hao Dong</a><sup>1,2,‡</sup>
</div>
</div>
</div>
<!-- Affiliations Section -->
<div style="text-align: center; margin-bottom: 15px;">
<h3 style="font-size: 1.2em; margin-bottom: 10px;"></h3>
<p style="margin: 0; font-size: 0.75em; line-height: 1.4;">
<sup>1</sup><a href="https://cfcs.pku.edu.cn/" class="affiliation-link" target="_blank">CFCS, School of Computer Science, Peking University</a>
<sup>2</sup><a href="https://www.pku.edu.cn" class="affiliation-link" target="_blank">PKU-Agibot Lab</a>
<sup>3</sup><a href="https://www.shlab.org.cn/" class="affiliation-link" target="_blank">Shanghai AI Laboratory</a>
</p>
</div>
<!-- Equal contribution note -->
<div style="text-align: center;">
<span class="author-note" style="font-size: 0.85em;"><sup>*</sup>Equal contribution</span>
<span class="author-note" style="font-size: 0.85em;"><sup>‡</sup>Corresponding author</span>
</div>
</div>
</div>
<script>
function toggleAnswer(answerId, element) {
const answerElement = document.getElementById(answerId);
const isVisible = answerElement.style.display === "block";
if (isVisible) {
answerElement.style.display = "none";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Click to view Ground Truth and MLLM\'s answer!</strong>';
} else {
answerElement.style.display = "block";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Hide answer</strong>';
if (window.innerWidth <= 768) {
// answerElement.scrollIntoView({ behavior: "smooth", block: "start" });
}
}
}
</script>
<script>
function toggleAnswer2(answerId, element) {
const answerElement = document.getElementById(answerId);
if (answerElement.style.visibility === "hidden" || answerElement.style.visibility === "") {
answerElement.style.transition = "none";
answerElement.style.visibility = "visible";
answerElement.style.opacity = "1";
answerElement.style.maxHeight = "500px";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Hide answer</strong>';
} else {
// Hide the answer instantly
answerElement.style.transition = "none";
answerElement.style.opacity = "0";
answerElement.style.maxHeight = "0";
answerElement.style.visibility = "hidden";
element.innerHTML = '<img src="static/img/icons/teaser.gif" style="width:1.5rem"> <strong>Click to view Ground Truth and MLLM\'s answer!</strong>';
}
}
</script>
</div>
<div class="icon-row">
<a href="#overall-video" class="icon-link">
<img src="static/img/icons/video.svg" alt="Vision Logo" class="icon">
Overall Introduction<br>Video
</a>
<a href="#NavSpace-Introduction" class="icon-link">
<img src="static/img/icons/benchmark.svg" alt="Visual Representation Logo" class="icon">
NavSpace<br>Introduction
</a>
<a href="#NavSpace-Evaluation" class="icon-link">
<img src="./static/img/icons/evaluation.svg" alt="Evaluation Logo" class="icon">
Evaluation<br>Protocol
</a>
<a href="#NavSpace-Leaderboard" class="icon-link">
<img src="/static/img/icons/leaderboard_black.svg" alt="Leaderboard Logo" class="icon">
NavSpace<br>Leaderboard
</a>
</div>
<p class="click-hint2" style="width: 85%;">
<img src="static/img/icons/click.gif" style="width: 1.5rem">
<strong>Click to jump to each section.</strong>
</p>
<hr>
<div id="overall-video" style="position: relative; margin-top: 50px; margin-bottom: 40px;">
<h2 class="text" style="margin-top:0px; margin-bottom:20px;">Overall Introduction Video</h2>
<div style="display: flex; justify-content: center; align-items: center; margin: 20px 0;">
<div style="
background: #87CEEB;
padding: 10px;
border-radius: 15px;
box-shadow: 0 10px 30px rgba(0,0,0,0.3);
max-width: 95%;
width: fit-content;
">
<video
controls
preload="metadata"
style="
width: 100%;
max-width: 1200px;
height: auto;
aspect-ratio: 16/9;
object-fit: contain;
border-radius: 10px;
box-shadow: 0 5px 15px rgba(0,0,0,0.2);
display: block;
background-color: #000;
"
poster="static/img/preview.png"
>
<source src="navspace_new.mp4" type="video/mp4">
您的浏览器不支持视频播放。
</video>
</div>
</div>
<p class="text" style="text-align: center; color: #666; font-style: italic; margin-top: 15px;">
NavSpace: How Navigation Agents Follow Spatial Intelligence Instructions
</p>
</div>
<d-figure id="fig-teaser">
<figure>
<img data-zoomable="" draggable="false" src="static/img/teaser.png" alt="Visual-Spatial Intelligence Teaser">
<figcaption>
<strong>Figure 1:</strong> (Left) NavSpace tasks. (Right) Evaluation results compare various models with baseline SNav.
</figcaption>
</figure>
</d-figure>
<div id='NavSpace-Introduction' class="vsi-benchmark">
<div id="sec:vsi-overview" class="sub-section">
<h1 class="text">NavSpace Introduction</h1>
<p class="text" align="justify">
<p class="text" align="justify">
<strong>Benchmark Overview:</strong>
We introduce the NavSpace benchmark, which
contains six task categories and 1,228 trajectory-instruction
pairs designed to probe the spatial intelligence of navigation
agents. On this benchmark, we comprehensively evaluate 22
navigation agents, including state-of-the-art navigation models
and multimodal large language models. The evaluation results
lift the veil on spatial intelligence in embodied navigation.
Furthermore, we propose SNav, a new spatially intelligent
navigation model. SNav outperforms existing navigation agents
on NavSpace and real robot tests, establishing a strong baseline
for future work.
</p>
<d-figure id="fig-task-demo" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/task-demo.png" alt="benchmark category">
<figcaption style="text-align: center; margin-top: 20px;">
<strong>Figure 2: Instruction Categories in NavSpace. </strong>
These six categories were determined based on the questionnaire survey results. Every navigation trajectory and instruction was collected manually from HM3D scene datasets through our designed platform.
</figcaption>
</figure>
</d-figure>
<d-figure id="fig-bench-stats">
<figure>
<img data-zoomable="" draggable="false" src="static/img/benchmark-stats.png" alt="benchmark category" style="width: 90%; height: auto; display: block; margin: 0 auto;">
<figcaption style="text-align: center; margin-top: 20px;">
<strong>Figure 3: Visualization of NavSpace Statistics. </strong>
</figcaption>
</figure>
</d-figure>
</div>
<template>
<div id="vsi-construct" class="sub-section">
<p class="text" align="justify"><strong>NavSpace Construction:</strong>
We designed a two-part survey to identify navigation instructions indicative of spatial intelligence. Participants first reviewed a definition of spatial intelligence and confirmed comprehension, then evaluated 17 candidate instruction types, selecting up to six. From 512 responses, 457 valid ones were retained, yielding six key categories: Vertical Perception, Precise Movement, Viewpoint Shifting, Spatial Relationship, Environment State, and Space Structure. To collect trajectories, we built a Habitat 3.0-based platform with HM3D scenes, integrating a front-end annotation webpage and simulator-linked backend. Annotators teleoperated agents in first-person view, recording RGB frames, actions, and coordinates. GPT-5 then analyzed trajectories, combining navigation data with sampled visual observations to generate candidate instructions, which annotators refined and finalized. For quality assurance, cross-validation required each instruction to be executed by a different annotator; unsuccessful cases were discarded and re-annotated. This pipeline ensured collected data were diverse in spatial reasoning and reliable for downstream evaluation.</p>
<d-figure id="fig-bench-pipeline" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/benchmark_construction.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>Figure 4: Construction pipeline of NavSpace.</strong> (1) Questionnaire Survey: identify navigation instruction types that best reflect spatial intelligence. (2) Trajectory Collection: teleoperate agents in simulation to record trajectories. (3) Instruction Annotation: generate instructions with large-model-assisted analysis. (4) Human Cross-Validation: review and validate instructions for correctness and executability.
</p>
</figcaption>
</figure>
</d-figure>
</div>
</template>
<div id="NavSpace-Evaluation" class="vsi-evaluation">
<h1 class="text">Evaluation on NavSpace</h1>
<script type="text/javascript" id="MathJax-script" async
src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js">
</script>
<p class="text" align="justify"><strong>Evaluation Environment and Metrics</strong>:
NavSpace takes Habitat 3.0 as the simulator to conduct the evaluation. Evaluation scenes are selected from the HM3D datasets. At each step, the agent
can only select one action. Following previous instruction
navigation benchmarks, we employ the following widely used evaluation metrics: Navigation Error (NE), Oracle Success Rate (OS), Success Rate (SR).
We conduct a comprehensive evaluation of existing multimodal large models and navigation models. These models can be categorized into the following five types:
</p>
<ul class="text" style="margin-left: -7.00em; padding-left: 0; list-style-position: inside;">
<li>Chance Level Baselines</li>
<li>Open-source MLLMs</li>
<li>Proprietary MLLMs</li>
<li>Lightweight Navigation Models</li>
<li>Navigation Large Models</li>
</ul>
<p class="text" align="justify"><strong>Performances on NavSpace</strong>:
NavSpace poses a major challenge for current models. Open-source MLLMs achieve less than 10% success, near chance level, while proprietary MLLMs perform slightly better, with GPT-5 leading but still under 20%. This indicates that existing MLLMs are far from reliable navigation agents for spatial intelligence tasks. Lightweight navigation models such as BEVBert and ETPNav also fail on NavSpace, whereas larger navigation models like NaVid and StreamVLN outperform GPT-5 and show preliminary spatial reasoning ability. Our proposed model, SNav, surpasses both state-of-the-art navigation models and advanced MLLMs, establishing a strong new baseline. Ablation results further demonstrate that its instruction-generation pipeline is key to enhancing spatial intelligence performance.
</p>
<d-figure id="fig-eval-bench" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/evaluation-bench.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>Table 1: Quantative performances on NavSpace.</strong> Bold color indicates the best result among all models</p>
</figcaption>
</figure>
</d-figure>
<p class="text" align="justify"><strong>Our SoTA Model——SNav</strong>:
The architecture of the SNav incorporates three fundamental
components: the Vision Encoder v(·), the Projector p(·),
and the Large Language Model (LLM) f(·). We follow the previous work to conduct navigation
finetuning through co-training with three tasks. These
tasks include Navigation Action Prediction, Trajectory-based
Instruction Generation, and General Multimodal Data Recall.
After this, we obtain the vanilla SNav model.
</p>
<p class="text" align="justify"><strong>Spatial Intelligence Enhancement</strong>:</p>
<ul class="text" style="margin-left: -7.00em; padding-left: 0; list-style-position: inside;">
<li><strong>Cross-floor Navigation</strong>: We identify R2R trajectories likely to cross floors by thresholding height differences between start and end. For each candidate, the agent follows a shortest-path planner in Habitat while recording RGB frames. A trajectory is labeled floor-crossing if GPT-5 detects stairs in ≥3 frames. Using methods from HOV-SG, we assign floor labels to start/end points and combine them with Habitat's floor count for vertical-space annotations. GPT-5 then restyles raw instructions (e.g., “Walk up the stairs ...”) into floor-specific ones (e.g., “Walk up to the third floor ...”).
</li>
<li><strong>Precise Movement</strong>: We sample start-goal pairs in MP3D scenes and compute shortest paths in Habitat. After filtering, trajectories of target length (20-60 steps) are retained, with discrete actions recorded (turn ±30°, move forward 0.25m, stop). Consecutive actions of the same type are merged into concise descriptions (e.g., “move forward 3 m, turn right 60°, move forward 2 m”). GPT-5 paraphrases these into natural-language instructions.
</li>
<li><strong>Environment State Inference</strong>: We extract start-end point pairs with navigation instructions from the R2R dataset and generate trajectories using a shortest-path planner, saving RGB frames along each path. GPT-5 is then queried with the first and last frames to infer three elements: observable objects, unobservable objects, and scene descriptions. After observing the structure of this category, we design five template categories that integrate multimodal observations with original instructions. Two representative patterns are: (1) Original_instruction; if [visible_object in last frame] then stop at [last-frame location], otherwise go to [scene description from first frame], and (2) If [fabricated_object in first frame] then stop, otherwise follow Original_instruction and stop at [last-frame location]. GPT-5 rewrites instructions according to these templates, producing high-quality training data for environment state inference.
</li>
<li><strong>Spatial Relationship</strong>: We filter R2R instructions with regular expressions for ordinal phrases (“first room/door”, “second room/door”, etc.) and detect multi-object relations using keywords like “between”, “along”, and “across”.
</li>
</ul>
<d-figure id="fig-enhancement" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/fig-enhancement.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>Figure 4: Spatial Intelligence Enhancement.</strong> </p>
</figcaption>
</figure>
</d-figure>
<p id="real-world-tests" class="text" align="justify"><strong>Real World Tests</strong>:
In the real-world test, we test our model SNav across three different environments:
office, campus, and outdoor environments. The test covers
five categories of spatially intelligent navigation instructions
(excluding vertical perception). Our experimental platform is
the AgiBot Lingxi D1 quadruped, which is equipped with
a monocular RGB camera and motion-control APIs.
</p>
<d-figure id="fig-real-world" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/real-world.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>Figure 5: Qualitative results from the real-world deployment of SNav.</strong> The evaluated instructions cover five categories
proposed in NavSpace. The test environment includes the office, the campus building, and the outdoor area.</p>
</figcaption>
</figure>
</d-figure>
<template>
<d-figure id="fig-real-world" >
<figure>
<img data-zoomable="" draggable="false" src="static/img/real-world-tests.png" alt="benchmark category">
<figcaption>
<p align="justify"><strong>TABLE 3: Quantitative Results of Real-World Experiments.</strong> </p>
</figcaption>
</figure>
</d-figure>
</template>
<!-- VSI-Bench Leaderboard Section -->
<div id="NavSpace-Leaderboard" class="vsi-leaderboard">
<h1 class="text">NavSpace Leaderboard</h1>
<div class="leaderboard-container">
<div class="leaderboard-header">
<p class="text" align="justify">
To include your model in the leaderboard, please email <a href="mailto:NavSpace@163.com"><strong>NavSpace@163.com</strong></a> with evaluation logs and setups.
</p>
</div>
<table class="leaderboard-table">
<thead>
<tr>
<th data-column-index="0">Model</th>
<th data-column-index="1">LLM Params</th>
<th data-column-index="2" style="display:none">Frames</th>
<th data-column-index="3" style="display:none">Date</th>
<th data-column-index="4">Avg Success Rate(%).</th>
<th data-column-index="5">Vertical Perception(SR)</th>
<th data-column-index="6">Precise Movement(SR)</th>
<th data-column-index="7">Viewpoint Shifting(SR)</th>
<th data-column-index="8">Spatial Relationship(SR)</th>
<th data-column-index="9">Environment State(SR)</th>
<th data-column-index="10">Space Structure(SR)</th>
<th data-column-index="11" style="display:none"></th>
<th data-column-index="12" style="display:none"></th>
</tr>
</thead>
<tbody id="leaderboard-body">
<tr class="section-divider">
<td colspan="13" class="section-header">Chance Level Baselines</td>
</tr>
<tr class="baseline-row">
<td>Chance-level (Random)</td>
<td>-</td>
<td>-</td>
<td>2025-09-25</td>
<td>4.2</td>
<td>4.3</td>
<td>1.0</td>
<td>3.9</td>
<td>4.2</td>
<td>6.0</td>
<td>5.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr class="baseline-row">
<td>Chance-level (Frequency)</td>
<td>-</td>
<td>-</td>
<td>2024-11-15</td>
<td>8.3</td>
<td>11.5</td>
<td>3.5</td>
<td>11.6</td>
<td>7.5</td>
<td>6.5</td>
<td>9.0</td>
<td>-</td>
<td>-</td>
</tr>
<tr class="section-divider">
<td colspan="13" class="section-header">Open-source MLLMs</td>
</tr>
<tr>
<td>LLaVA-Video</td>
<td>7B</td>
<td>64</td>
<td>2025-09-26</td>
<td>6.7</td>
<td>7.7</td>
<td>3.5</td>
<td>6.8</td>
<td>11.3</td>
<td>6.5</td>
<td>4.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>GLM-4.1V-Thinking </td>
<td>9B</td>
<td>64</td>
<td>2025-09-26</td>
<td>6.5</td>
<td>7.7</td>
<td>2.0</td>
<td>8.2</td>
<td>11.3</td>
<td>7.0</td>
<td>3.0</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>GLM-4.5V </td>
<td>106B</td>
<td>64</td>
<td>2025-09-26</td>
<td>7.6</td>
<td>7.7</td>
<td>2.5</td>
<td>7.2</td>
<td>9.4</td>
<td>12.0</td>
<td>6.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>Qwen2.5-VL</td>
<td>7B</td>
<td>64</td>
<td>2025-09-26</td>
<td>7.5</td>
<td>6.3</td>
<td>2.5</td>
<td>7.7</td>
<td>9.4</td>
<td>8.5</td>
<td>10.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>Qwen2.5-VL</td>
<td>72B</td>
<td>64</td>
<td>2025-09-26</td>
<td>7.0</td>
<td>9.1</td>
<td>3.0</td>
<td>5.3</td>
<td>6.1</td>
<td>8.5</td>
<td>10.0</td>
<td>-</td>
<td>-</td>
</tr>
<tr class="section-divider">
<td colspan="13" class="section-header">Proprietary MLLMs</td>
</tr>
<tr>
<td>GPT-4o </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>7.7</td>
<td>10.1</td>
<td>4.0</td>
<td>3.9</td>
<td>9.9</td>
<td>8.5</td>
<td>9.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>GPT-5 Mini </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>10.9</td>
<td>15.4</td>
<td>4.0</td>
<td>5.8</td>
<td>12.3</td>
<td>14.0</td>
<td>14.0</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>GPT-5 </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>14.2</td>
<td>18.3</td>
<td>3.0</td>
<td>12.6</td>
<td>17.5</td>
<td>17.5</td>
<td>16.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>Gemini 2.5 Flash </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>6.5</td>
<td>3.8</td>
<td>4.0</td>
<td>4.8</td>
<td>7.5</td>
<td>11.5</td>
<td>7.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>Gemini 2.5 Pro </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>11.2</td>
<td>23.6</td>
<td>4.0</td>
<td>9.2</td>
<td>7.1</td>
<td>13.0</td>
<td>10.0</td>
<td>-</td>
<td>-</td>
</tr>
<tr class="section-divider">
<td colspan="13" class="section-header">Lightweight Nav Models</td>
</tr>
<tr>
<td>Seq2Seq </td>
<td>-</td>
<td>16</td>
<td>2025-09-26</td>
<td>0.7</td>
<td>1.0</td>
<td>0.0</td>
<td>0.0</td>
<td>1.4</td>
<td>1.5</td>
<td>0.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>CMA </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>1.8</td>
<td>0.5</td>
<td>0.0</td>
<td>1.4</td>
<td>2.8</td>
<td>5.5</td>
<td>0.5</td>
<td>-</td>
<td>-</td>
</tr>
<tr>
<td>HPN+DN </td>
<td>-</td>
<td>32</td>
<td>2025-09-26</td>
<td>8.5</td>
<td>8.7</td>
<td>3.5</td>