-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathindex.html
4399 lines (4317 loc) · 349 KB
/
index.html
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>
<head>
<title>Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making</title>
<!-- consider to add our icon here -->
<!-- <link rel="icon" href="" type="image/icon type"> -->
<meta name="viewport" content="width=device-width, initial-scale=1">
<script src="https://cdn.jsdelivr.net/npm/chart.js@4.4.0/dist/chart.umd.min.js"></script>
<script src="https://cdn.jsdelivr.net/npm/chartjs-plugin-datalabels@2.0.0"></script>
<script
src="https://cdn.jsdelivr.net/npm/chartjs-plugin-annotation@3.0.1/dist/chartjs-plugin-annotation.min.js"></script>
<link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro" rel="stylesheet">
<link rel="stylesheet" href="website/css/bulma.min.css">
<link rel="stylesheet" href="website/css/bulma-carousel.min.css">
<link rel="stylesheet" href="website/css/bulma-slider.min.css">
<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/jpswalsh/academicons@1/css/academicons.min.css">
<link rel="stylesheet" href="website/css/fontawesome.all.min.css">
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
<script src="./website/javascript/bulma-carousel.min.js"></script>
<script src="./website/javascript/bulma-slider.min.js"></script>
<link href="https://cdn.jsdelivr.net/npm/bootstrap@5.1.3/dist/css/bootstrap.min.css" rel="stylesheet"
integrity="sha384-1BmE4kWBq78iYhFldvKuhfTAU6auU8tT94WrHftjDbrCEXSU1oBoqyl2QvZ6jIW3" crossorigin="anonymous">
<script src="https://cdn.jsdelivr.net/npm/bootstrap@5.1.3/dist/js/bootstrap.bundle.min.js"
integrity="sha384-ka7Sk0Gln4gmtz2MlQnikT1wXgYsOg+OMhuP+IlRH9sENBO0LRn5q+8nbTov4+1p"
crossorigin="anonymous"></script>
<link href="https://unpkg.com/tabulator-tables@5.5.2/dist/css/tabulator_bootstrap4.min.css" rel="stylesheet">
<script type="text/javascript" src="https://unpkg.com/tabulator-tables@5.5.2/dist/js/tabulator.min.js"></script>
<script defer src="website/javascript/fontawesome.all.min.js"></script>
<!-- <script src="website/javascript/peity-vanilla.js"></script> -->
<!-- below we load some js scripts -->
<script src="website/javascript/benchmark_table.js" type="module"></script>
<link rel="stylesheet" href="website/css/index.css">
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-C7GJ4FYMY9"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag() { dataLayer.push(arguments); }
gtag('js', new Date());
gtag('config', 'G-C7GJ4FYMY9');
</script>
<!-- MathJax script -->
<script type="text/javascript" async
src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/MathJax.js?config=TeX-MML-AM_CHTML">
</script>
<script type="text/javascript">
MathJax.Hub.Config({
tex2jax: {inlineMath: [['$','$'], ['\\(','\\)']]}
});
</script>
<noscript>
<p><img alt="Clicky" width="1" height="1" src="//in.getclicky.com/101339888ns.gif" /></p>
</noscript>
<script>
document.addEventListener('DOMContentLoaded', function () {
var toggles = document.querySelectorAll('.toggle-section');
toggles.forEach(function(toggle) {
toggle.addEventListener('click', function() {
var content = document.getElementById(toggle.getAttribute('aria-controls'));
var toggleIcon = toggle.children[1].children[0];
content.classList.toggle('is-active');
if (content.classList.contains('is-active')) {
toggleIcon.style.transition = 'transform 0.3s ease';
toggleIcon.style.transform = 'rotate(180deg)';
} else {
toggleIcon.style.transition = 'transform 0.3s ease';
toggleIcon.style.transform = 'rotate(0deg)';
}
});
});
});
</script>
<style>
.collapse-content {
display: none;
margin-top: 10px;
}
.collapse-content.is-active {
display: block;
}
/* .toggle-section .icon.is-small {
transition: transform 0.3s ease;
} */
/* .toggle-section .fa-angle-up {
transform: rotate(180deg);
} */
</style>
</head>
<body>
<section class="hero">
<div class="hero-body">
<div class="container is-max-desktop">
<div class="columns is-centered">
<div class="column has-text-centered">
<h1 class="title is-1 publication-title is-bold">
<!-- <img src="website/img/mint-leaf-logo.png" alt="logo" width="40" height="40" /> -->
Embodied Agent Interface: Benchmarking LLMs for Embodied Decision Making
</h1>
<p style="color: darkred; font-weight: bold;">Evaluate your model with a single line of code!</p>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://limanling.github.io/">Manling Li</a><sup>1, 2†</sup>,</span>
<span class="author-block">
<a href="https://www.linkedin.com/in/shiyu-zhao-1124a0266/">Shiyu Zhao</a><sup>1,†</sup>,</span>
<span class="author-block">
<a href="https://qinengwang-aiden.github.io/">Qineng Wang</a><sup>1, 2†</sup>,
</span>
<span class="author-block">
<a href="https://jameskrw.github.io/">Kangrui Wang</a><sup>1, 2†</sup>,
</span>
<span class="author-block">
<a href="https://bryanzhou008.github.io/">Yu Zhou</a><sup>1,†</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://www.linkedin.com/in/sanjana-srivastava5/">Sanjana Srivastava</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://www.cemgokmen.com/">Cem Gokmen</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="https://profiles.stanford.edu/tonyhlee">Tony Lee</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="http://www.cs.columbia.edu/~lierranli/">Li Erran Li</a><sup>3</sup>,
</span>
<span class="author-block">
<a href="https://ai.stanford.edu/~zharu/">Ruohan Zhang</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="http://weiyuliu.com/">Weiyu Liu</a><sup>1</sup>,
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block">
<a href="https://cs.stanford.edu/~pliang/">Percy Liang</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="http://vision.stanford.edu/feifeili/">Li Fei-Fei</a><sup>1</sup>,
</span>
<span class="author-block">
<a href="http://jiayuanm.com/">Jiayuan Mao</a><sup>4</sup>,
</span>
<span class="author-block">
<a href="https://jiajunwu.com/">Jiajun Wu</a><sup>1</sup>
</span>
</div>
<div class="is-size-5 publication-authors">
<span class="author-block"><sup>1</sup>Stanford University,</span>
<span class="author-block"><sup>2</sup>Northwestern University,</span>
<span class="author-block"><sup>3</sup>Amazon,</span>
<span class="author-block"><sup>4</sup>MIT</span>
</div>
<div class="'is-size-5 publication-authors">
<span class="author-block"><sup>†</sup>Equal contribution</span>
</div>
<div class="'is-size-4 publication-authors">
<span class="author-block" style="color: darkred; font-weight: bold;">NeurIPS 2024 D&B (Oral)</span>
</div>
<div class="column has-text-centered">
<div class="publication-links">
<!-- PDF Link. -->
<!-- <span class="link-block">
<a class="btn btn-outline-dark"
role="button">
<i class="fas fa-file-pdf"></i>
<span> Paper (Coming Soon)</span>
</a>
</span> -->
<span class="link-block">
<a href="https://arxiv.org/abs/2410.07166" class="btn btn-outline-dark"
role="button">
<span class="icon">
<i class="ai ai-arxiv"></i>
</span>
<span>arXiv</span>
</a>
</span>
<!-- Code Link. -->
<span class="link-block">
<a href="https://github.com/embodied-agent-eval/embodied-agent-eval" class="btn btn-outline-dark" role="button">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Code</span>
</a>
</span>
<!-- Dataset Link. -->
<span class="link-block">
<a href="https://huggingface.co/datasets/Inevitablevalor/EmbodiedAgentInterface" class="btn btn-outline-dark" role="button" style="display: inline-flex; align-items: center;">
<span class="icon" style="display: inline-flex; align-items: center;">
<img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" alt="Hugging Face" style="width: 20px; height: 18px; margin-right: 5px;">
</span>
<span>Dataset</span>
</a>
</span>
<!-- Dockerhub Link. -->
<span class="link-block">
<a href="https://hub.docker.com/r/jameskrw/eai-eval" class="btn btn-outline-dark" role="button">
<span class="icon"><i class="fab fa-docker"></i></span>
<span>Docker</span>
</a>
</span>
<!-- Behavior Pypi Python Package Link. -->
<span class="link-block">
<a href="https://pypi.org/project/eai-eval/" class="btn btn-outline-dark" role="button">
<span class="icon"><i class="fab fa-python"></i></span>
<span>EAI</span>
</a>
</span>
<!-- Virtualhome Pypi Python Package Link. -->
<!-- <span class="link-block">
<a href="https://pypi.org/project/virtualhome-eval/" class="btn btn-outline-dark" role="button">
<span class="icon"><i class="fab fa-python"></i></span>
<span>VirtualHome Eval</span>
</a>
</span> -->
<!-- Docs Link. -->
<span class="link-block">
<a href="https://embodied-agent-eval.readthedocs.io/en/latest/#" class="btn btn-outline-dark" role="button">
<span class="icon">
<i class="fa fa-book"></i>
</span>
<span>Docs</span>
</a>
</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Add the video element here -->
<!-- <video autoplay muted loop playsinline style="width: 100%; height: auto;">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/data/eagent.mp4" type="video/mp4">
Your browser does not support the video tag.
</video> -->
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="hero-body">
<video id="teaser" autoplay muted loop playsinline controls width="100%">
<!-- <source src="./website/videos/eai-video-v2.mov" type="video/quicktime"> -->
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-0-overview.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</section>
<section class="section">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column">
<h3 class="title is-5">Goal Interpretation</h3>
<video autoplay muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-1-goal.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="column">
<h3 class="title is-5">Subgoal Decomposition</h3>
<video autoplay muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-2-subgoal.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="column">
<h3 class="title is-5">Action Sequencing</h3>
<video autoplay muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-3-action.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
<div class="column">
<h3 class="title is-5">Transition Modeling</h3>
<video autoplay muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-4-transition.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
<div class="columns is-centered has-text-centered">
<div class="column">
<div class="column">
<!-- Raw video link button
<a href="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/eai.mp4" target="_blank" class="button is-light">
<span class="icon">
<i class="fas fa-download"></i>
</span>
<span>Download Full Video</span>
</a>
-->
<!-- Toggle button with consistent style -->
<button class="button is-light toggle-section" aria-controls="full_demo_video">
<span class="icon">
<i class="fas fa-play"></i>
</span>
<span>Watch Full Demo</span>
<span class="icon">
<i class="fas fa-angle-down" style="margin-left: 5px;"></i>
</span>
</button>
<!-- Collapsible video container -->
<div id="full_demo_video" class="collapse-content" style="margin-top: 1rem;">
<video id="demo" controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/eai.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</div>
</div>
</div>
</div>
</section>
<section class="section" id="abstract">
<div class="container is-max-desktop">
<!-- Abstract. -->
<div class="columns is-centered has-text-centered">
<div class="column is-four-fifths">
<h2 class="title is-2">Abstract</h2>
<div class="content has-text-justified">
<p>
<b>Problem:</b> We aim to evaluate Large Language Models (LLMs) for embodied decision making. While a significant body of work has been leveraging LLMs for decision making in embodied environments, we still lack a systematic understanding of their performances, because they are usually applied in different domains for different purposes, and built based on different inputs and outputs. Furthermore, existing evaluations tend to rely solely on a final success rate, making it difficult to pinpoint what ability is missing in LLMs and where the problem lies, which in turn, blocks embodied agents from leveraging LLMs effectively and selectively.
</p>
<p>
<b>Method:</b> To address these limitations, we propose a generalized interface (<b>Embodied Agent Interface</b>) that supports the formalization of various types of tasks and input-output specifications of LLM-based modules. Specifically, it allows us to unify <b>1)</b> a broad set of embodied decision making tasks involving both state and temporally extended goals, <b>2)</b> four commonly-used LLM-based modules for decision making: goal interpretation, subgoal decomposition, action sequencing, and transition modeling, and <b>3)</b> a collection of fine-grained metrics which break down evaluation into various types of errors, such as hallucination errors, affordance errors, various types of planning errors, etc.
</p>
<p>
<b>Conclusion:</b> Overall, our benchmark offers a comprehensive and systematic assessment of LLMs' performance for different subtasks, pinpointing the strengths and weaknesses in LLM-powered embodied AI systems, and providing insights for effective and selective use of LLMs in embodied decision making.
</p>
</div>
<figure>
<img src="website/img/teaser.png" alt="Embodied agent interface overview." class="EAgent_overview_image"/>
<figcaption style="font-style: italic;">
<b>Figure 1:</b> <b>Embodied Agent Interface</b> unifies a broad set of tasks involving both state and temporally extended goals and four LLM-based modules for decision making.
</figcaption>
</figure>
</div>
</div>
<!--/ Abstract. -->
</div>
</section>
<section class="hero teaser">
<div class="container is-max-desktop">
<div class="column is-centered">
<h2 class="title is-2" style="text-align: center;">Embodied Agent Interface</h2>
<br>
<div class="content has-text-justified">
<p>
In our Embodied Agent Interface, we propose a set of ability modules to evaluate LLMs for embodied decision making. The four ability modules are: Goal Interpretation, Subgoal Decomposition, Action Sequencing, and Transition Modeling. We provide a detailed description of each module below.
</p>
<h3 class="title is-4">Ability Module 1: Goal Interpretation</h3>
<!-- <video id="goal-interpretation-video" muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-1-goal.mp4" type="video/mp4">
Your browser does not support the video tag.
</video> -->
<p>
Goal Interpretation aims to ground the natural language instruction to the environment representations of objects, states, relations, and actions. For example, the task instruction "Use the rag to clean the trays, the bowl, and the refrigerator. When you are done, leave the rag next to the sink..." can be grounded to specific objects with IDs, such as fridge (ID: 97), tray (ID: 1), bowl (ID: 1), rag (ID: 0), and sink (ID: 82). Note that a simple natural language description can be grounded into a set of multiple goal conditions (object state and relation).
</p>
<h3 class="title is-4">Ability Module 2: Subgoal Decomposition</h3>
<!-- <video id="subgoal-video" muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-2-subgoal.mp4" type="video/mp4">
Your browser does not support the video tag.
</video> -->
<p>
Subgoal Decomposition generates a sequence of states, where each state can be a set of objects and their states. Here, we highlight the important states, such as the transitions between a sequence of next_to(rag.0, sink.82), toggled_on(sink.82), soaked(rag.0), toggled_off(sink.82), open(fridge.97), not_stained(fridge.97). To achieve these state transitions, we can use a high-level planner such as BFS to search for the Action Sequences that achieve these state transitions. We obtain the following action sequence: RIGHT_GRASP(rag.0), RIGHT_PLACE_NEXTTO(sink.82), TOGGLE_ON(sink.82), SOAK(rag.0), TOGGLE_OFF(sink.82), OPEN(fridge.97), CLEAN(fridge.97). Note that multiple actions may be required to achieve a single one-step state transition. For example, to perform the state transition next_to(rag.0, sink.82) → toggled_on(sink.82), we need two actions RIGHT_GRASP(rag.0), RIGHT_PLACE_NEXTTO(sink.82). See <a href="#EAgent_taxtonomy_example">Figure 2</a> for the input and output formulation.
</p>
<figure id="EAgent_taxtonomy_example">
<img src="website/img/taxonomy-ability.png" alt="Embodied agent interface taxonomy example." class="EAgent_taxtonomy_image" style="width: 75%;"/>
<figcaption>
<b>Figure 2:</b> The input and output formulation of four ability modules for <b>Embodied Agent Interface</b>.
</figcaption>
</figure>
<h3 class="title is-4">Ability Module 3: Action Sequencing</h3>
<!-- <video id="action-sequencing-video" muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-3-action.mp4" type="video/mp4">
Your browser does not support the video tag.
</video> -->
<p>
Action Sequences are essential to achieve the state transitions identified in Subgoal Decomposition. For example, a successful execution of the action sequence RIGHT_GRASP(rag.0), RIGHT_PLACE_NEXTTO(sink.82), TOGGLE_ON(sink.82), SOAK(rag.0), TOGGLE_OFF(sink.82), OPEN(fridge.97), CLEAN(fridge.97) is shown in <a href="#EAgent_excution_example">Figure 3</a>.
</p>
<h3 class="title is-4">Ability Module 4: Transition Modeling</h3>
<!-- <video id="transition-modeling-video" muted loop playsinline controls width="100%">
<source src="https://github.com/embodied-agent-eval/embodied-agent-eval.github.io/raw/main/website/videos/eai-4-transition.mp4" type="video/mp4">
Your browser does not support the video tag.
</video> -->
<p>
Transition Modeling serves as the low-level controller to guide the simulator in performing state transitions from preconditions to post-effects. For example, in cleaning task, the input is the operator name soak, and the preconditions are three states: holding (?obj1), next_to (?sink ?agent), and toggled_on (?sink). The post effect after executing SOAK is soaked (?obj1).
</p>
<figure id="EAgent_excution_example">
<img src="website/img/excution_example.png" alt="Example of successful execution in Embodied Agent Interface." class="EAgent_excution_image" style="width: 75%;"/>
<figcaption>
<b>Figure 3:</b> An example of successful execution in <b>Embodied Agent Interface</b>.
</figcaption>
</figure>
</div>
</div>
<div class="hero-body">
<!-- <h2 class="subtitle">
<b>Embodied Agent Interface</b> aims to tackle the following challenges in evaluating LLMs for building embodied decision-making agents: (1) Standardization of goal specifications. (2) Standardization of modules and interfaces. (3) Broad coverage of evaluation and fine-grained metrics.
</h2> -->
<h2 class="title is-2" style="text-align: center;">Dataset Viewer</h2>
<iframe
src="https://huggingface.co/datasets/Inevitablevalor/EmbodiedAgentInterface/embed/viewer/default/behavior"
frameborder="0"
width="100%"
height="560px"
></iframe>
<br>
<br>
<h2 class="title is-2" style="text-align: center;">Leaderboard</h2>
<ul class="nav nav-tabs" id="myTab" role="tablist">
<li class="nav-item" role="presentation">
<button class="nav-link active" id="main-results-tab" data-bs-toggle="tab"
data-bs-target="#benchmark-table-content" type="button" role="tab"
aria-controls="main-results-tab" aria-selected="true">BEHAVIOR</button>
</li>
<li class="nav-item" role="presentation">
<button class="nav-link" id="eurus-code-table-tab" data-bs-toggle="tab"
data-bs-target="#eurus-code-table-content" type="button" role="tab"
aria-controls="eurus-code-table-tab" aria-selected="false">VirtualHome</button>
</li>
</ul>
<div class="tab-content" id="myTabContent">
<div class="tab-pane fade show active" id="benchmark-table-content" role="tabpanel"
aria-labelledby="benchmark-table-content">
<div id="behavior-benchmark-main-table"></div>
</div>
<div class="tab-pane fade" id="eurus-code-table-content" role="tabpanel"
aria-labelledby="eurus-code-table-content">
<div id="virtualhome-benchmark-main-table"></div>
</div>
</div>
<br>
<h2 class="title is-2" style="text-align: center;">Empirical Findings</h2>
<div class="content has-text-justified">
<ol>
<li><strong>Goal Interpretation:</strong>
<ul>
<li>LLMs struggle to translate natural language instructions into grounded states.</li>
<li>Common errors include generating intermediate goals and omitting spatial relationship goals.</li>
<li>Gemini 1.5 Pro has the highest goal interpretation performance, while Claude-3 Opus excels in goal retrieval rate.</li>
<li>Proprietary LLMs make fewer grammar errors compared to open-source LLMs.</li>
</ul>
<div class="collapsible-section">
<button class="button is-fullwidth toggle-section" aria-controls="goal_interpretation_table">
<span>View full results of goal interpretation</span>
<span class="icon is-small">
<i class="fas fa-angle-down" aria-hidden="true"></i>
</span>
</button>
<div id="goal_interpretation_table" class="collapse-content">
<table class="table is-striped is-hoverable" id="goal_interpretation" style="width: 100%; table-layout: fixed; font-size: 0.9em;">
<caption style="caption-side: top; text-align: center; color: black; font-style: italic;">
<b>Table:</b> All goal evaluation results (%) for goal interpretation
</caption>
<thead>
<tr>
<th rowspan="4" style="width: 15%;">Model Name</th>
<th colspan="24" style="text-align: center; background-color: #EED3D9; width: 85%;">Goal Interpretation</th>
</tr>
<tr>
<th colspan="6" style="text-align: center;">State</th>
<th colspan="6" style="text-align: center;">Spatial</th>
<th colspan="6" style="text-align: center;">Action</th>
<th colspan="6" style="text-align: center;">Overall</th>
</tr>
<tr>
<th colspan="2" style="text-align: center;">Precision</th>
<th colspan="2" style="text-align: center;">Recall</th>
<th colspan="2" style="text-align: center;">F1</th>
<th colspan="2" style="text-align: center;">Precision</th>
<th colspan="2" style="text-align: center;">Recall</th>
<th colspan="2" style="text-align: center;">F1</th>
<th colspan="2" style="text-align: center;">Precision</th>
<th colspan="2" style="text-align: center;">Recall</th>
<th colspan="2" style="text-align: center;">F1</th>
<th colspan="2" style="text-align: center;">Precision</th>
<th colspan="2" style="text-align: center;">Recall</th>
<th colspan="2" style="text-align: center;">F1</th>
</tr>
<tr>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
<th style="text-align: center;">V</th>
<th style="text-align: center;">B</th>
</tr>
</thead>
<tbody>
<tr>
<td>Claude-3 Haiku</td>
<td>21.8</td>
<td>22.8</td>
<td>58.9</td>
<td>93.5</td>
<td>31.8</td>
<td>36.7</td>
<td>24.2</td>
<td>64.5</td>
<td>50.8</td>
<td>64.6</td>
<td>32.8</td>
<td>64.6</td>
<td>12.2</td>
<td>-</td>
<td><b>95.7</b></td>
<td>-</td>
<td>21.6</td>
<td>-</td>
<td>18.0</td>
<td>41.5</td>
<td>63.2</td>
<td>71.2</td>
<td>28.0</td>
<td>52.5</td>
</tr>
<tr>
<td>Claude-3 Sonnet</td>
<td>23.3</td>
<td>36.8</td>
<td>57.1</td>
<td>88.9</td>
<td>33.1</td>
<td>52.0</td>
<td>26.6</td>
<td>76.2</td>
<td>53.0</td>
<td>79.8</td>
<td>35.5</td>
<td>77.9</td>
<td>12.4</td>
<td>-</td>
<td>85.8</td>
<td>-</td>
<td>21.7</td>
<td>-</td>
<td>19.3</td>
<td>60.2</td>
<td>61.5</td>
<td>81.9</td>
<td>29.4</td>
<td>69.4</td>
</tr>
<tr>
<td>Claude-3 Opus</td>
<td>27.0</td>
<td>72.6</td>
<td>66.9</td>
<td>93.5</td>
<td>38.5</td>
<td>81.7</td>
<td>22.6</td>
<td>75.2</td>
<td>46.8</td>
<td>79.2</td>
<td>30.5</td>
<td>77.1</td>
<td>14.5</td>
<td>-</td>
<td>92.6</td>
<td>-</td>
<td>25.1</td>
<td>-</td>
<td>20.7</td>
<td>72.2</td>
<td>65.0</td>
<td>82.5</td>
<td>31.4</td>
<td>77.0</td>
</tr>
<tr>
<td>Claude-3.5 Sonnet</td>
<td>25.3</td>
<td>74.0</td>
<td>60.9</td>
<td>94.8</td>
<td>35.8</td>
<td>83.1</td>
<td>31.1</td>
<td><b>84.4</b></td>
<td><b>63.8</b></td>
<td>81.3</td>
<td>41.8</td>
<td><b>82.9</b></td>
<td>14.0</td>
<td>-</td>
<td><b>98.8</b></td>
<td>-</td>
<td>24.5</td>
<td>-</td>
<td>21.7</td>
<td><b>81.1</b></td>
<td><b>69.6</b></td>
<td>84.4</td>
<td>33.0</td>
<td><b>82.7</b></td>
</tr>
<tr>
<td>Cohere Command R</td>
<td><b>51.1</b></td>
<td>7.7</td>
<td><b>69.6</b></td>
<td>31.4</td>
<td><b>58.9</b></td>
<td>12.4</td>
<td>34.5</td>
<td>56.8</td>
<td>21.3</td>
<td>55.0</td>
<td>26.3</td>
<td>55.9</td>
<td>3.6</td>
<td>-</td>
<td>38.9</td>
<td>-</td>
<td>6.5</td>
<td>-</td>
<td>27.4</td>
<td>28.2</td>
<td>55.7</td>
<td>49.6</td>
<td>36.7</td>
<td>36.0</td>
</tr>
<tr>
<td>Cohere Command R+</td>
<td>20.9</td>
<td>23.3</td>
<td>52.0</td>
<td>79.1</td>
<td>29.8</td>
<td>36.0</td>
<td>17.9</td>
<td>66.7</td>
<td>15.2</td>
<td>61.5</td>
<td>16.4</td>
<td>64.0</td>
<td>10.4</td>
<td>-</td>
<td>82.6</td>
<td>-</td>
<td>18.5</td>
<td>-</td>
<td>14.9</td>
<td>42.0</td>
<td>44.5</td>
<td>65.5</td>
<td>22.4</td>
<td>51.2</td>
</tr>
<tr>
<td>Gemini 1.0 Pro</td>
<td>25.3</td>
<td>27.4</td>
<td>57.9</td>
<td>81.1</td>
<td>34.9</td>
<td>41.0</td>
<td>17.0</td>
<td>75.2</td>
<td>20.6</td>
<td>70.4</td>
<td>18.6</td>
<td>72.7</td>
<td>9.9</td>
<td>-</td>
<td>68.7</td>
<td>-</td>
<td>17.2</td>
<td>-</td>
<td>16.2</td>
<td>51.0</td>
<td>45.2</td>
<td>72.8</td>
<td>23.8</td>
<td>60.0</td>
</tr>
<tr>
<td>Gemini 1.5 Flash</td>
<td>23.6</td>
<td>55.8</td>
<td>57.9</td>
<td>94.1</td>
<td>33.5</td>
<td>70.1</td>
<td>19.8</td>
<td>76.6</td>
<td>21.1</td>
<td>76.7</td>
<td>20.5</td>
<td>76.7</td>
<td>13.5</td>
<td>-</td>
<td>90.1</td>
<td>-</td>
<td>23.5</td>
<td>-</td>
<td>18.2</td>
<td>69.7</td>
<td>50.8</td>
<td>80.7</td>
<td>26.8</td>
<td>74.8</td>
</tr>
<tr>
<td>Gemini 1.5 Pro</td>
<td>45.4</td>
<td><b>94.0</b></td>
<td>49.1</td>
<td>92.8</td>
<td>47.2</td>
<td><b>93.4</b></td>
<td>40.0</td>
<td>74.4</td>
<td>9.7</td>
<td>76.7</td>
<td>15.6</td>
<td>75.6</td>
<td><b>26.8</b></td>
<td>-</td>
<td>80.9</td>
<td>-</td>
<td><b>40.3</b></td>
<td>-</td>
<td><b>35.2</b></td>
<td>78.8</td>
<td>41.1</td>
<td>80.4</td>
<td>37.9</td>
<td>79.6</td>
</tr>
<tr>
<td>GPT-3.5-turbo</td>
<td>22.4</td>
<td>52.0</td>
<td>50.0</td>
<td>66.7</td>
<td>30.9</td>
<td>58.5</td>
<td>8.5</td>
<td>51.5</td>
<td>18.8</td>
<td>46.9</td>
<td>11.7</td>
<td>49.1</td>
<td>15.2</td>
<td>-</td>
<td>60.5</td>
<td>-</td>
<td>24.4</td>
<td>-</td>
<td>15.7</td>
<td>49.5</td>
<td>40.5</td>
<td>51.4</td>
<td>22.7</td>
<td>50.4</td>
</tr>
<tr>
<td>GPT-4-turbo</td>
<td>28.6</td>
<td>70.4</td>
<td>58.5</td>
<td>86.9</td>
<td>38.4</td>
<td>77.8</td>
<td>24.7</td>
<td>77.5</td>
<td>32.9</td>
<td>76.4</td>
<td>28.2</td>
<td>76.9</td>
<td>19.0</td>
<td>-</td>
<td>82.1</td>
<td>-</td>
<td>30.9</td>
<td>-</td>
<td>24.0</td>
<td>75.6</td>
<td>53.8</td>
<td>78.8</td>
<td>33.2</td>
<td>77.2</td>
</tr>
<tr>
<td>GPT-4o</td>
<td>29.0</td>
<td>67.1</td>
<td>60.0</td>
<td>94.8</td>
<td>39.1</td>
<td>78.6</td>
<td>31.5</td>
<td>81.1</td>
<td>43.6</td>
<td>78.5</td>
<td>36.6</td>
<td>79.8</td>
<td>20.5</td>
<td>-</td>
<td>85.8</td>
<td>-</td>
<td>33.1</td>
<td>-</td>
<td>26.4</td>
<td>76.5</td>
<td>59.1</td>
<td>82.2</td>
<td>36.5</td>
<td>79.2</td>
</tr>
<tr>
<td>Llama 3 8B Instruct</td>
<td>21.7</td>
<td>17.3</td>
<td>54.4</td>
<td>80.4</td>
<td>31.0</td>
<td>28.4</td>
<td>14.0</td>
<td>51.4</td>
<td>7.4</td>
<td>20.8</td>
<td>9.7</td>
<td>29.6</td>
<td>11.1</td>
<td>-</td>
<td>79.4</td>
<td>-</td>
<td>19.4</td>
<td>-</td>
<td>15.5</td>
<td>24.1</td>
<td>41.9</td>
<td>34.3</td>
<td>22.6</td>
<td>28.3</td>
</tr>
<tr>
<td>Llama 3 70B Instruct</td>
<td>23.9</td>
<td>69.5</td>
<td>61.2</td>
<td><b>95.4</b></td>
<td>34.3</td>
<td>80.4</td>
<td>22.6</td>
<td>70.0</td>
<td>37.5</td>
<td>73.3</td>
<td>28.2</td>
<td>71.6</td>
<td>11.2</td>
<td>-</td>
<td>88.8</td>
<td>-</td>
<td>19.8</td>
<td>-</td>
<td>17.5</td>
<td>64.7</td>
<td>58.0</td>
<td>78.3</td>
<td>26.9</td>
<td>70.9</td>
</tr>
<tr>
<td>Mistral Large</td>
<td>23.6</td>
<td>63.5</td>
<td>59.1</td>
<td>92.2</td>
<td>32.8</td>
<td>75.2</td>
<td>23.7</td>
<td>75.1</td>
<td>40.3</td>
<td>76.2</td>
<td>29.8</td>
<td>75.6</td>
<td>11.2</td>
<td>-</td>
<td>84.0</td>
<td>-</td>
<td>19.7</td>
<td>-</td>
<td>17.5</td>
<td>69.6</td>
<td>57.1</td>
<td>79.8</td>
<td>26.8</td>
<td>74.3</td>
</tr>
<tr>
<td>Mixtral 8x22B MoE</td>
<td>23.6</td>
<td>22.9</td>
<td>56.9</td>
<td>83.7</td>
<td>33.4</td>
<td>36.0</td>
<td>22.2</td>
<td>70.7</td>
<td>36.3</td>
<td>67.7</td>
<td>27.5</td>
<td>69.2</td>
<td>11.2</td>
<td>-</td>
<td>94.8</td>
<td>-</td>
<td>20.0</td>
<td>-</td>
<td>17.4</td>
<td>44.4</td>
<td>56.2</td>
<td>71.3</td>
<td>26.6</td>
<td>54.7</td>
</tr>
<tr>
<td>o1-mini</td>
<td>26.3</td>
<td>63.8</td>
<td>58.6</td>
<td>90.8</td>
<td>36.3</td>
<td>74.9</td>
<td>30.4</td>
<td>77.3</td>
<td>39.9</td>
<td>76.5</td>
<td>34.5</td>
<td>76.9</td>
<td>13.5</td>