EscapeCraft/index.html at main · THUNLP-MT/EscapeCraft · GitHub

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
<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <link rel="stylesheet" type="text/css" href="static/css/table.css">
  <!-- Meta tags for social media banners, these should be filled in appropriatly as they are your "business card" -->
  <!-- Replace the content tag with appropriate information -->
  <meta name="description" content="DESCRIPTION META TAG">
  <meta property="og:title" content="SOCIAL MEDIA TITLE TAG"/>
  <meta property="og:description" content="SOCIAL MEDIA DESCRIPTION TAG TAG"/>
  <meta property="og:url" content="URL OF THE WEBSITE"/>
  <!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X630-->
  <meta property="og:image" content="static/image/your_banner_image.png" />
  <meta property="og:image:width" content="1200"/>
  <meta property="og:image:height" content="630"/>


  <meta name="twitter:title" content="TWITTER BANNER TITLE META TAG">
  <meta name="twitter:description" content="TWITTER BANNER DESCRIPTION META TAG">
  <!-- Path to banner image, should be in the path listed below. Optimal dimenssions are 1200X600-->
  <meta name="twitter:image" content="static/images/your_twitter_banner_image.png">
  <meta name="twitter:card" content="summary_large_image">
  <!-- Keywords for your paper to be indexed by-->
  <meta name="keywords" content="KEYWORDS SHOULD BE PLACED HERE">
  <meta name="viewport" content="width=device-width, initial-scale=1">


  <title>MM-Escape</title>
  <link rel="icon" type="image/x-icon" href="static\images\escape.png">
  <link href="https://fonts.googleapis.com/css?family=Google+Sans|Noto+Sans|Castoro"
  rel="stylesheet">

  <link rel="stylesheet" href="static/css/bulma.min.css">
  <link rel="stylesheet" href="static/css/bulma-carousel.min.css">
  <link rel="stylesheet" href="static/css/bulma-slider.min.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">
  <link rel="stylesheet" href="static/css/index.css">

  <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.js"></script>
  <script src="https://documentcloud.adobe.com/view-sdk/main.js"></script>
  <script defer src="static/js/fontawesome.all.min.js"></script>
  <script src="static/js/bulma-carousel.min.js"></script>
  <script src="static/js/bulma-slider.min.js"></script>
  <script src="static/js/index.js"></script>
  <style>
    /* 水印样式 */
    body::after {
        content: "EscapeCraft Group from THUMT";
        position: fixed;
        top: 50%;
        left: 30%; /* 水平居中 */
        transform: translate(-50%, -50%); /* 调整为完全居中 */
        font-size: 3em;
        color: rgba(0, 0, 0, 0.1);
        font-weight: bold;
        z-index: 9999;
        pointer-events: none;
        white-space: nowrap;
        text-align: center;
        transform: rotate(-30deg);
    }
  </style>
</head>
<body>

  <div class="title-bar">
    <div class="buttons">
      <a href="#abstract" class="button is-small is-light" style = "font-size: 18px; font-family: 'Google Sans', sans-serif;">Abstract</a>
      <a href="#introduction" class="button is-small is-light" style = "font-size: 18px; font-family: 'Google Sans', sans-serif;">Introduction</a>
      <a href="#game settings" class="button is-small is-light" style = "font-size: 18px; font-family: 'Google Sans', sans-serif;">Settings</a>
      <a href="#analysis" class="button is-small is-light" style = "font-size: 18px; font-family: 'Google Sans', sans-serif;">Analysis</a>
      <a href="#results" class="button is-small is-light" style = "font-size: 18px; font-family: 'Google Sans', sans-serif;">Results</a>

    </div>
  </div>
  <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">How Do Multimodal Large Language Models Handle Complex Multimodal Reasoning? Placing Them in An Extensible Escape Game</h1>
            <div class="is-size-5 publication-authors">
              <!-- Paper authors -->
              <span class="author-block">
                <a href="https://openreview.net/profile?id=~Ziyue_Wang4" target="_blank">Ziyue Wang</a><sup style="color: rebeccapurple;">1</sup>,</span>
                <span class="author-block">
                  <a href="https://openreview.net/profile?id=~Yurui_Dong1" target="_blank">Yurui Dong</a><sup style="color: rgb(205, 140, 70);">2</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Fuwen_Luo1" target="_blank">Fuwen Luo</a><sup style="color: rebeccapurple;">1</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Minyuan_Ruan1" target="_blank">Minyuan Ruan</a><sup style="color: rebeccapurple;">1</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Zhili_Cheng1" target="_blank">Zhili Cheng</a><sup style="color: rebeccapurple;">1</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Chi_Chen1" target="_blank">Chi Chen</a><sup style="color: rebeccapurple;">1</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Peng_Li2" target="_blank">Peng Li</a><sup style="color: rebeccapurple;">1</sup>,</span>
                  </span>
                  <span class="author-block">
                    <a href="https://openreview.net/profile?id=~Yang_Liu19" target="_blank">Yang Liu</a><sup style="color: rebeccapurple;">1</sup></span>
                  </span>
                  </div>

                  <div class="is-size-5 publication-authors">
                    <span class="author-block"><sup style="color: rebeccapurple;">1 </sup>Tsinghua University<br><sup style="color: rgb(205, 140, 70);">2 </sup style="color: rgb(205, 140, 70);">Fudan University</span>
                  </div>

                  <div class="column has-text-centered">
                    <div class="publication-links">
                         <!-- Arxiv PDF link -->
                      <span class="link-block">
                        <a href="https://arxiv.org/pdf/2503.10042" target="_blank"
                        class="external-link button is-normal is-rounded is-dark">
                        <span class="icon">
                          <i class="fas fa-file-pdf"></i>
                        </span>
                        <span>Paper</span>
                      </a>
                    </span>

                    <!-- Supplementary PDF link
                    <span class="link-block">
                      <a href="static/pdfs/supplementary_material.pdf" target="_blank"
                      class="external-link button is-normal is-rounded is-dark">
                      <span class="icon">
                        <i class="fas fa-file-pdf"></i>
                      </span>
                      <span>Supplementary</span>
                    </a>
                  </span> -->

                  <!-- Github link -->
                  <span class="link-block">
                    <a href="https://github.com/THUNLP-MT/EscapeCraft" target="_blank"
                    class="external-link button is-normal is-rounded is-dark">
                    <span class="icon">
                      <i class="fab fa-github"></i>
                    </span>
                    <span>Code</span>
                  </a>
                </span>

                <!-- ArXiv abstract Link -->
                <span class="link-block">
                  <a href="https://arxiv.org/abs/2503.10042" target="_blank"
                  class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                    <i class="ai ai-arxiv"></i>
                  </span>
                  <span>arXiv</span>
                </a>
              </span>
            </div>
          </div>
        </div>
      </div>
    </div>
  </div>
</section>

<video width="640" height="360" controls style="margin-bottom: 40px;">
  <source src="static/video/video.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" style="display: flex; justify-content: center; align-items: center; flex-direction: column;">
      <img
        src="static/images/teaser_v6_00.png"
        alt="Teaser Image"
        style="max-width: 140%; height: auto; display: block; border-radius: 10px; box-shadow: 0px 4px 10px rgba(0, 0, 0, 0.1);">

      <div style="max-width: 130%; text-align: center; font-size: 18px; line-height: 1.6; font-family: 'Times New Roman', serif; margin-top: 20px;">
        Illustration of our proposed room escape environment <strong>EscapeCraft</strong>, which allows us to generate customized room scenes (left) and define groundtruth reasoning path of tasks (right). Based on EscapeCraft, we create <strong>MM-Escape benchmark</strong>, targeting at evaluating both the task completion performance and the entire multimodal reasoning process of MLLMs.
      </div>
    </div>
  </div>
</section>


<!-- End teaser video -->

<!-- Paper abstract -->
<section id="abstract" class="section hero is-light">
  <div class="container is-fluid">
    <div class="columns is-vcentered">

      <!-- 左侧：竖向 Abstract -->
      <div class="column is-2 has-text-centered">
        <h2 class="vertical-title">Abstract</h2>
      </div>

      <!-- 右侧：正文内容 -->
      <div class="column is-10">
        <div class="content has-text-justified">
          <p>
            <div style="font-family: 'Times New Roman', serif; font-size: 20px;">
              The rapid advancing of Multimodal Large Language Models (MLLMs) has spurred interest in complex multimodal reasoning tasks in the real-world and virtual environment, which require coordinating multiple abilities, including visual perception, visual reasoning, spatial awareness, and target deduction.
              However, existing evaluations primarily assess the final task completion, often degrading assessments to isolated abilities such as visual grounding and visual question answering. Less attention is given to comprehensively and quantitatively analyzing reasoning process in multimodal environments, which is crucial for understanding model behaviors and underlying reasoning mechanisms beyond merely task success.
              To address this, we introduce MM-Escape, an extensible benchmark for investigating multimodal reasoning, inspired by real-world escape games. MM-Escape emphasizes intermediate model behaviors alongside final task completion. To achieve this, we develop EscapeCraft, a customizable and open environment that enables models to engage in free-form exploration for assessing multimodal reasoning.
              Extensive experiments show that MLLMs, regardless of scale, can successfully complete the simplest room escape tasks, with some exhibiting human-like exploration strategies. Yet, performance dramatically drops as task difficulty increases. Moreover, we observe that performance bottlenecks vary across models, revealing distinct failure modes and limitations in their multimodal reasoning abilities, such as repetitive trajectories without adaptive exploration, getting stuck in corners due to poor visual spatial awareness, and ineffective use of acquired props, such as the key.
              We hope our work sheds light on new challenges in multimodal reasoning, and uncovers potential improvements in MLLMs capabilities.
            </div>
          </p>
        </div>
      </div>

    </div>
  </div>
</section>
<!-- End paper abstract -->

<!-- introduction -->
<section id="introduction" >
  <div class="container is-fluid">
    <div class="columns is-vcentered">

      <div class="column is-10">
        <div class="content has-text-justified" style="padding-bottom: 20px;">
          <p>
            <div style="font-family: 'Times New Roman', serif; font-size: 20px;padding-left:50px;padding-top: 15px;padding-right: 25px;">
              The rapid development of Large Language Models (LLMs) and Multimodal Large Language Models (MLLMs) has driven advancements in various applications, requiring strong multimodal reasoning abilities such as visual perception, spatial awareness, and tool utilization. However, existing evaluation methods primarily focus on task completion rather than analyzing the entire reasoning process, limiting insights into model capabilities.<br>
              To address this, we introduce <strong>MM-Escape</strong>, a benchmark inspired by real-world escape games, designed to assess complex multimodal reasoning. At its core, we develop <strong>EscapeCraft</strong>, a customizable open environment that enables models to engage in free-form exploration through room escape tasks. This allows for a comprehensive evaluation of reasoning abilities beyond simple task completion, focusing on intermediate behaviors and decision-making.<br>
              Our findings reveal that while MLLMs show promising reasoning abilities, performance drops significantly as task complexity increases, exposing distinct failure patterns. These insights emphasize the need for more comprehensive analysis and improvements in multimodal reasoning.
            </div>
          </p>
        </div>
      </div>
      <div class="column is-2 has-text-centered">
        <h2 class="vertical-title">Introduction</h2>
      </div>
    </div>
  </div>
</section>
<!-- End of Introduction -->

<!-- Methodology -->
<section id="game settings"  class="section hero is-light">
  <div class="container is-fluid">

    <!-- 标题 -->
    <h2 class="title has-text-centered">Game Settings</h2>

    <div class="content has-text-justified" style="font-family: 'Times New Roman', serif; font-size: 20px;padding-left: 20px;">
      <p>We introduce an automatic reasoning chain generation procedure by configuring the <i>Prop Chain</i>. In detail, this is a singly linked list representing the ordered sequence of items and interactions required to complete the game. Each node in the chain corresponds to an interactive element, such as a key, a locked box, or a note with a password, where the tail node represents the exit point of the game. To construct a complete escape game setting, we annotate the links between nodes in the prop chain to define the ways to obtain different props (such as unlimited acquisition or requiring a key to open, etc.) and their inclusion relationships (for example, a key can be placed in a box)</p>
    </div>

    <!-- Difficulty + 图片分栏 -->
    <div class="columns is-centered">

      <!-- 左侧 Difficulty 文字（缩小宽度） -->
      <div class="column is-8">
        <div class="content has-text-justified" style="font-family: 'Times New Roman', serif; font-size: 20px;padding-left: 20px;">
          <p><b>Difficulty-1:</b> The simplest one-hop reasoning path where no props are needed to unlock the door. Models can exit by locating the door and interacting with it directly.</p>

          <p><b>Difficulty-2:</b> A two-hop reasoning path requiring an additional key or password compared to Difficulty-1. Models should search for the key or password and interact with it to unlock the door.</p>

          <p><b>Difficulty-3:</b> A three-hop reasoning path requiring both a password and a key, with one additional hop to Difficulty-2. This level challenges models with spatial reasoning, visual search, and prop utilization.</p>

          <p>Since the prop chain can grow infinitely, our difficulty levels are inherently extendable. Moerover, the type of questions or tasks in each reasoning hop are customizable and interchangeable, further enhancing the difficulty and flexibility of MM-Escape. We also explore some extended settings that incorporate with other tasks, such as embodied QA and visual logical reasoning</p>
        </div>
      </div>

      <div class="column is-4 has-text-centered">
        <figure class="image">
          <img src="static/images/levels_extend_00.png" alt="Game Levels" style="width: 100%; height: auto;padding-left: 10px;">
        </figure>
      </div>

    </div>

  </div>
</section>

<!-- End of Methodology -->

<!-- beginning of analysis -->
<section id = "analysis">
  <div class="container is-fluid">
    <h2 class="title is-3" style="padding: 30px;">Analysis</h2>
    <div class="columns is-centered">
      <div class="column is-10">
        <div class="content has-text-justified" style="font-family: 'Times New Roman', serif; font-size: 20px;padding-left: 20px;">
          <p style="border: 2px dashed black; border-radius: 10px; padding: 10px;">
            <strong>Takeaways</strong>
            <br><strong>&bull; </strong><i>Distinct Human-Like Behavioral Patterns:</i> Models exhibit unique behaviors in the escape task. Gemini tends to remain in a fixed location at the start, scanning its surroundings before acting, while GPT-4o first observes a wider range for a more global understanding.
            <br><strong>&bull; </strong><i>Diverse Observation Strategies: </i>Gemini frequently adopts a downward-facing view to inspect objects like tables and chairs, whereas GPT-4o predominantly relies on a front-facing view.
            <br><strong>&bull; </strong><i>Common Failure Modes:</i> Models struggle with movement and interaction failures. GPT-4o tends to repeat trajectories, while Gemini and Claude get stuck in confined areas. Phi-3 and Qwen-VL fail at action combinations, leading to imprecise object interactions.
            <br><strong>&bull; </strong><i>Adherence to Long-Term Goals: </i>Most models focus on locating the exit and key objects, but Phi-3 consistently fails to recognize doors, unlike other models that engage with doors upon detection.
        </p>
            <strong>Analysis of Entire Path</strong><br>
            The analysis focuses on three key aspects: (1) the number of steps required to obtain core props, (2) steps needed to exit after acquiring the key or password, and (3) the relationship between grab success rate (GSR) and escape success.
            Efficiency in Obtaining Key Props: GPT-4o demonstrates the best performance in locating and acquiring essential props, followed by Gemini. Claude performs well in higher difficulty levels but with a lower escape rate.
            Efficiency in Exiting: Gemini performs well in moderate difficulty but struggles in complex environments, where GPT-4o benefits from prior memory and spatial understanding, leading to more efficient escapes.
            Grab Success Rate (GSR) and Escape Success: A higher GSR correlates with better escape performance. GPT-4o and Claude 3.5 maintain relatively stable GSRs, while Qwen and Llama struggle due to weaker environmental perception and interaction capabilities.<br>
            <strong>Extensibility of EscapeCraft</strong><br>
            An extended case study evaluates model adaptability to different room configurations. In a scenario where the password is displayed as a numerical pattern instead of written text:
            GPT-4o quickly recognizes and uses it when the pattern is near the door but struggles when the pattern is farther away, revealing limitations in long-term and spatial reasoning.
            Gemini fails to recognize the pattern entirely, instead resorting to exhaustive room searches.
            This highlights the importance of enhancing models' ability to interpret environmental cues dynamically.<br>
            <strong>Post-game Debriefing</strong><br>
            After escaping, models are tasked with recalling and reconstructing their escape process. Only GPT-4o and Gemini-1.5-Pro are evaluated due to their higher escape rates. Findings show that:
            Models prioritize key actions (e.g., obtaining the password) but neglect background context, suggesting a trade-off between efficiency and holistic understanding.
            Future improvements should focus on enhancing memory and reasoning abilities to better capture contextual details beyond task completion.</p>
          </div>
          <div id="results-carousel" class="carousel results-carousel">
            <div class="item">
             <!-- Your image here -->
             <img src="static\images\example_v3_00.png" alt="MY ALT TEXT"/>
             <h2 class="subtitle has-text-centered" style="font-size: 15px;font-family: 'Gill Sans', 'Gill Sans MT', Calibri, 'Trebuchet MS', sans-serif;padding-top: 18px;text-align: left;">
              An example of success escape path from Gemini-1.5-pro. For the first four steps, it rotates, without moving, to observe the
              surroundings, and then starts moving towards the direct on the observed tables in the following steps. For step 5 and 6, we zoom in to the
              key to make it clearer. The key is distorted because the view of model is tilting downwards to observe objects on the table. From step 10
              to 14, the model is already close to the door, and is slowly rotating and tilting to locate the door.
             </h2>
           </div>
           <div class="item">
             <!-- Your image here -->
             <img src="static\images\analysis-result_v3_00.png" alt="MY ALT TEXT"/>
             <h2 class="subtitle has-text-centered" style="font-size: 15px;font-family: 'Gill Sans', 'Gill Sans MT', Calibri, 'Trebuchet MS', sans-serif;padding-top: 18px;text-align: left;">
              Analysis on the grabbing behaviors with respect to the escape rate and the visibility of the exit at initial location.
             </h2>
           </div>
           <div class="item">
             <!-- Your image here -->
             <img src="static\images\finalcase_00.png" alt="MY ALT TEXT"/>
             <h2 class="subtitle has-text-centered" style="font-size: 15px;font-family: 'Gill Sans', 'Gill Sans MT', Calibri, 'Trebuchet MS', sans-serif;padding-top: 18px;text-align: left;">
              The extended scenario where the required password is displayed via a numerical pattern on the wall, rather than explicitly written
              on notes. GPT-4o completes reading it at once and exits within five steps, while Gemini struggles to repetitively search the room. Moving
              the pattern away from the door further challenges GPT-4o, leading to a failure of escaping.
            </h2>
          </div>
       </div>
      </div>
    </div>
  </div>
</section>

<!-- Benchmark result -->
<section id = "results">
<head>
  <meta charset="UTF-8">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  <title>Experiment Results</title>
  <script>
      function showTable(tableId, button) {
          // 隐藏所有表格
          document.getElementById('single-room-table').classList.add('hidden');
          document.getElementById('multi-room-table').classList.add('hidden');
          document.getElementById(tableId).classList.remove('hidden');

          // 切换按钮高亮状态
          document.querySelectorAll('.button-group button').forEach(btn => btn.classList.remove('active'));
          button.classList.add('active');
      }

      // 页面加载时默认选中 Single Room
      window.onload = function () {
          document.getElementById('single-room-btn').classList.add('active');
      };
  </script>
</head>

  <h2 class="title is-3" style="padding: 30px;">Experiment Results</h2>

  <!-- 按钮组 -->
  <div class="button-group">
      <button id="single-room-btn" onclick="showTable('single-room-table', this)">Single Room</button>
      <button id="multi-room-btn" onclick="showTable('multi-room-table', this)">Multi-Room</button>
  </div>
    <!-- Single Room Table -->
    <table id="single-room-table">
      <tr>
          <th rowspan="2">Models</th>
          <th colspan="4">Difficulty-1</th>
          <th colspan="5">Difficulty-2</th>
          <th colspan="5">Difficulty-3</th>
          <th rowspan="2">AVG ER (%)↑</th>
      </tr>
      <tr>
          <th>ER (%)↑</th>
          <th>Steps↓</th>
          <th>Grab SR (%)↑</th>
          <th>Grab Ratio</th>
          <th>ER (%)↑</th>
          <th>Prop (%)↑</th>
          <th>Steps↓</th>
          <th>Grab SR (%)↑</th>
          <th>Grab Ratio</th>
          <th>ER (%)↑</th>
          <th>Prop (%)↑</th>
          <th>Steps↓</th>
          <th>Grab SR (%)↑</th>
          <th>Grab Ratio</th>
      </tr>
      <tr class="group-header" style="color:grey;">
          <td>Human</td>
          <td>100.00</td>
          <td>5.73</td>
          <td>95.45</td>
          <td>0.19</td>
          <td>100.00</td>
          <td>100.00</td>
          <td>13.64</td>
          <td>81.81</td>
          <td>0.19</td>
          <td>100.00</td>
          <td>100.00</td>
          <td>21.45</td>
          <td>75.45</td>
          <td>0.19</td>
          <td>100.00</td>
      </tr>
      <tr class="highlight-green">
          <td>GPT-4o</td>
          <td class="bold">100.00</td>
          <td class="bold">11.27</td>
          <td>37.82</td>
          <td>0.42</td>
          <td class="bold">72.73</td>
          <td>81.82</td>
          <td class="bold">36.73</td>
          <td class="bold">36.73</td>
          <td>0.26</td>
          <td class="bold">71.36</td>
          <td class="bold">90.00</td>
          <td class="bold">50.19</td>
          <td class="bold">31.36</td>
          <td>0.35</td>
          <td>81.36</td>
      </tr>
      <tr class="highlight-yellow">
          <td>Gemini-1.5-pro</td>
          <td>81.82</td>
          <td>21.18</td>
          <td class="bold">49.18</td>
          <td>0.39</td>
          <td>54.55</td>
          <td class="bold">90.91</td>
          <td>47.82</td>
          <td>14.89</td>
          <td>0.44</td>
          <td>46.82</td>
          <td>74.49</td>
          <td>73.18</td>
          <td>10.43</td>
          <td>0.48</td>
          <td>61.06</td>
      </tr>
      <tr class="highlight-yellow">
          <td>Claude 3.5 Sonnet</td>
          <td>72.73</td>
          <td>22.09</td>
          <td>30.64</td>
          <td>0.36</td>
          <td>45.45</td>
          <td>54.55</td>
          <td>57.45</td>
          <td>20.64</td>
          <td>0.17</td>
          <td>39.61</td>
          <td>54.83</td>
          <td>82.36</td>
          <td>16.21</td>
          <td>0.22</td>
          <td>52.60</td>
      </tr>
      <tr class="highlight-yellow">
        <td>Doubao 1.5 Pro</td>
        <td>91.91</td>
        <td>16.27</td>
        <td>44.68</td>
        <td>0.27</td>
        <td>45.45</td>
        <td>54.55</td>
        <td>63.18</td>
        <td>13.63</td>
        <td>0.25</td>
        <td>9.52</td>
        <td>33.33</td>
        <td>93.19</td>
        <td>6.76</td>
        <td>0.26</td>
        <td>48.96</td>
    </tr>
      <tr class="highlight-orange">
          <td>Llama-3.2-11b-vision</td>
          <td>63.64</td>
          <td>23.55</td>
          <td>31.36</td>
          <td>0.35</td>
          <td>0.00</td>
          <td>27.27</td>
          <td>75.00</td>
          <td>3.16</td>
          <td>0.44</td>
          <td>0.00</td>
          <td>27.27</td>
          <td>100.00</td>
          <td>3.55</td>
          <td>0.32</td>
          <td>21.21</td>
      </tr>
      <tr class="highlight-orange">
          <td>Qwen-VL-Max</td>
          <td>18.18</td>
          <td>42.64</td>
          <td>11.36</td>
          <td>0.05</td>
          <td>0.00</td>
          <td>27.27</td>
          <td>75.00</td>
          <td>3.51</td>
          <td>0.15</td>
          <td>9.09</td>
          <td>18.18</td>
          <td>94.18</td>
          <td>2.72</td>
          <td>0.31</td>
          <td>9.09</td>
      </tr>
      <tr class="highlight-orange">
        <td>Phi-3-vision-128k</td>
        <td>0.00</td>
        <td>50.00</td>
        <td>0.00</td>
        <td>0.01</td>
        <td>0.00</td>
        <td>0.00</td>
        <td>75.00</td>
        <td>0.00</td>
        <td>0.02</td>
        <td>0.00</td>
        <td>0.00</td>
        <td>100.00</td>
        <td>0.00</td>
        <td>0.01</td>
        <td>0.00</td>
    </tr>
    <tr class="highlight-orange">
      <td>GLM-4v Flash</td>
      <td>0.00</td>
      <td>50.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>75.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>100.00</td>
      <td>0.00</td>
      <td>0.00</td>
      <td>0.00</td>
  </tr>
  </table>

    <!-- Multi-Room Table -->
    <table id="multi-room-table" class="hidden">
      <thead>
          <tr>
              <th rowspan="2">Models</th>
              <th colspan="4">Difficulty-1 & Difficulty-1</th>
              <th colspan="5">Difficulty-1 & Difficulty-2</th>
              <th colspan="5">Difficulty-2 & Difficulty-2</th>
          </tr>
          <tr>
              <th>ER (%)↑</th>
              <th>Steps↓</th>
              <th>Grab SR (%)↑</th>
              <th>Grab Ratio</th>
              <th>ER (%)↑</th>
              <th>Prop (%)↑</th>
              <th>Steps↓</th>
              <th>Grab SR (%)↑</th>
              <th>Grab Ratio</th>
              <th>ER (%)↑</th>
              <th>Prop (%)↑</th>
              <th>Steps↓</th>
              <th>Grab SR (%)↑</th>
              <th>Grab Ratio</th>
          </tr>
      </thead>
      <tbody>
          <tr>
              <td>GPT-4o</td>
              <td class="bold">75.00</td>
              <td class="bold">35.50</td>
              <td>34.25</td>
              <td>0.32</td>
              <td class="bold">90.00</td>
              <td>100.00</td>
              <td class="bold">34.90</td>
              <td class="bold">35.52</td>
              <td>0.31</td>
              <td class="bold">70.00</td>
              <td class="bold">80.00</td>
              <td class="bold">39.50</td>
              <td class="bold">42.32</td>
              <td>0.37</td>
          </tr>
          <tr>
              <td>Gemini-1.5-pro</td>
              <td>22.22</td>
              <td>40.22</td>
              <td class="bold">22.89</td>
              <td>0.38</td>
              <td>40.00</td>
              <td class="bold">50.00</td>
              <td>56.60</td>
              <td>16.79</td>
              <td>0.05</td>
              <td>60.00</td>
              <td>80.00</td>
              <td>60.00</td>
              <td>22.71</td>
              <td>0.34</td>
          </tr>
          <tr>
              <td>Claude 3.5 Sonnet</td>
              <td>22.22</td>
              <td>45.22</td>
              <td>10.62</td>
              <td>0.08</td>
              <td>20.00</td>
              <td>20.00</td>
              <td>71.90</td>
              <td>6.75</td>
              <td>0.09</td>
              <td>10.00</td>
              <td>10.00</td>
              <td>80.00</td>
              <td>23.20</td>
              <td>0.06</td>
          </tr>
          <tr>
              <td>Llama-3.2-11b-vision</td>
              <td>55.56</td>
              <td>31.00</td>
              <td>36.25</td>
              <td>0.36</td>
              <td>10.00</td>
              <td>60.00</td>
              <td>66.40</td>
              <td>4.40</td>
              <td>0.40</td>
              <td>10.00</td>
              <td>40.00</td>
              <td>76.80</td>
              <td>27.00</td>
              <td>0.19</td>
          </tr>
          <tr>
              <td>Qwen-VL-Max</td>
              <td>22.22</td>
              <td>40.33</td>
              <td>12.96</td>
              <td>0.16</td>
              <td>30.00</td>
              <td>50.00</td>
              <td>57.70</td>
              <td>42.30</td>
              <td>0.28</td>
              <td>0.00</td>
              <td>10.00</td>
              <td>80.00</td>
              <td>23.66</td>
              <td>0.32</td>
          </tr>
      </tbody>
  </table>
</section>

</html>

<!-- End result -->


<!--BibTex citation -->
  <section class="section" id="BibTeX">
    <div class="container is-max-desktop content">
      <h2 class="title">BibTeX</h2>
      <pre style = "text-align: left;"><code>@misc{wang2025multimodallargelanguagemodels,
      title={How Do Multimodal Large Language Models Handle Complex Multimodal Reasoning? Placing Them in An Extensible Escape Game},
      author={Ziyue Wang and Yurui Dong and Fuwen Luo and Minyuan Ruan and Zhili Cheng and Chi Chen and Peng Li and Yang Liu},
      year={2025},
      eprint={2503.10042},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.10042},
  }</code></pre>
    </div>
</section>
<!--End BibTex citation -->


  <footer class="footer">
  <div class="container">
    <div class="columns is-centered">
      <div class="column is-8">
        <div class="content">

          <p>
            This page was built using the <a href="https://github.com/eliahuhorwitz/Academic-project-page-template" target="_blank">Academic Project Page Template</a> which was adopted from the <a href="https://nerfies.github.io" target="_blank">Nerfies</a> project page.
            You are free to borrow the source code of this website, we just ask that you link back to this page in the footer. <br> This website is licensed under a <a rel="license"  href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">Creative
            Commons Attribution-ShareAlike 4.0 International License</a>.
          </p>

        </div>
      </div>
    </div>
  </div>
</footer>

<!-- Statcounter tracking code -->

<!-- You can add a tracker to track page visits by creating an account at statcounter.com -->

    <!-- End of Statcounter Code -->

  </body>
  </html>