SkillComposition/index.html at main · ComputationalRobotics/SkillComposition · 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
<!DOCTYPE html>
<html>
<head>
  <meta charset="utf-8">
  <meta name="description"
      content="ICRA 2026 paper on skill composition for robot learning. We propose scene graph-based atomic skills to improve compositional generalization in long-horizon robotic manipulation tasks. Includes benchmark, code, and real-world experiments.">
  <meta name="keywords" content="
  skill composition robotics,
  robot skill composition,
  scene graph robot learning,
  scene graph representation robotics,
  long-horizon robot manipulation,
  compositional generalization robotics,
  robot manipulation learning,
  imitation learning for robotics,
  diffusion policy robotics,
  graph neural networks robotics,
  vision-language model robotics,
  VLM task planning,
  robot planning and control,
  generalist robots,
  atomic skills robotics,
  ICRA 2026 robotics paper,
  robot learning benchmark,
  robotics real-world manipulation,
  Harvard Computational Robotics
  ">
  <meta name="viewport" content="width=device-width, initial-scale=1">
  <meta property="og:title" content="Skill Composition for Robot Learning | ICRA 2026">
  <meta property="og:description" content="Scene graph-based atomic skills for compositional generalization in long-horizon robotic manipulation. Paper, code, and benchmark available.">
  <meta property="og:type" content="website">
  <meta property="og:url" content="https://computationalrobotics.seas.harvard.edu/SkillComposition/">
  <meta property="og:image" content="https://computationalrobotics.seas.harvard.edu/SkillComposition/static/images/mainfigure_v2.png">
  <meta name="twitter:card" content="summary_large_image">
  <meta name="twitter:title" content="Skill Composition for Robot Learning (ICRA 2026)">
  <meta name="twitter:description" content="Scene graph-based skill composition for long-horizon manipulation. Code and benchmark released.">
  <meta name="twitter:image" content="https://computationalrobotics.seas.harvard.edu/SkillComposition/static/images/mainfigure_v2.png">
  <title>Skill Composition for Robot Learning | Scene Graph Atomic Skills (ICRA 2026)</title>

  <!-- Bootstrap CSS -->
  <link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.0/dist/css/bootstrap.min.css" rel="stylesheet">

  <!-- Bootstrap JS (with Popper.js) -->
  <script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.0/dist/js/bootstrap.bundle.min.js"></script>

  <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">
  <link rel="icon" href="https://seas.harvard.edu/themes/custom/seas/favicon.ico">

  <script src="https://ajax.googleapis.com/ajax/libs/jquery/3.5.1/jquery.min.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> -->
  <script type="application/ld+json">
  {
    "@context": "https://schema.org",
    "@type": "ScholarlyArticle",
    "headline": "Scene Graph-based Atomic Skills for Skill Composition in Robot Learning",
    "name": "Compose by Focus",
    "author": [
      {
        "@type": "Person",
        "name": "Han Qi"
      },
      {
        "@type": "Person",
        "name": "Changhe Chen"
      },
      {
        "@type": "Person",
        "name": "Heng Yang"
      }
    ],
    "publisher": {
      "@type": "Organization",
      "name": "ICRA 2026"
    },
    "url": "https://computationalrobotics.seas.harvard.edu/SkillComposition/"
  }
  </script>
</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">
            Compose by Focus: Scene Graph-based Atomic Skills
          </h1>
          <h2 class="title is-3 publication-title">
              ICRA 2026
          </h2>
          <div class="is-size-5 publication-authors">
            <span class="author-block">
              <a href="https://han20192019.github.io/">Han Qi</a><sup>1</sup>,
            </span>
            <span class="author-block">
              <a href="https://aisenginn.github.io/">Changhe Chen</a><sup>2</sup>,
            </span>
            <span class="author-block">
              <a href="https://hankyang.seas.harvard.edu/">Heng Yang</a><sup>1</sup>
            </span>

          </div>

          <div class="is-size-5 publication-authors">
            <span class="author-block">
              <sup>1</sup>School of Engineering and Applied Sciences, Harvard University
            </span><br>
            <span class="author-block">
              <sup>2</sup>Robotics Department, University of Michigan
            </span><br>
            <!-- <span class="author-block">
              <sup>&dagger;</sup>Work done during visit at the
              <a href="https://computationalrobotics.seas.harvard.edu/">
                Harvard Computational Robotics Lab
              </a>
            </span> -->
          </div>

          <div class="column has-text-centered">
            <div class="publication-links">
              <!-- PDF Link. -->
              <!-- <span class="link-block">
                <a href="https://arxiv.org/pdf/2410.05063"
                   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> -->
              <span class="link-block">
                <a href="https://arxiv.org/abs/2509.16053/"
                   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>
              <!-- Video Link. -->
              <!-- <span class="link-block">
                <a href="https://www.youtube.com/watch?v=MrKrnHhk8IA"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fab fa-youtube"></i>
                  </span>
                  <span>Video</span>
                </a>
              </span> -->
              <!-- Code Link. -->
              <span class="link-block">
                <!-- https://github.com/han20192019/skill-composition-benchmark -->
                <a href="https://github.com/han20192019/skill-composition-benchmark"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="fab fa-github"></i>
                  </span>
                  <span>Benchmark</span>
                </a>
              </span>

              <span class="link-block">
                <!-- https://github.com/han20192019/skill-composition-code -->
                <a href="https://github.com/han20192019/skill-composition-code"
                   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>

              <span class="link-block">
              <a href="https://colab.research.google.com/drive/1SC5LtujSxPg8VEdq5rJdLHosLFLJ_a81?usp=sharing"
                class="external-link button is-normal is-rounded is-warning">
                <span class="icon">
                    <i class="fas fa-play"></i>
                </span>
                <span>Colab</span>
              </a>
            </span>
              <!-- Dataset Link. -->
              <!-- <span class="link-block">
                <a href="https://github.com/google/nerfies/releases/tag/0.1"
                   class="external-link button is-normal is-rounded is-dark">
                  <span class="icon">
                      <i class="far fa-images"></i>
                  </span>
                  <span>Data</span>
                </a>
              </span>
            </div> -->
          </div>
        </div>
      </div>
    </div>
  </div>
</section>


<section class="section">
  <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-3">Abstract</h2>
        <div class="content has-text-justified">
          <p>
            A key requirement for generalist robots is compositional generalization—the ability to combine atomic skills to solve complex, long-horizon tasks. While prior work has primarily focused on synthesizing a planner that sequences pre-learned skills, robust execution of the individual skills themselves remains challenging, as visuomotor policies often fail under distribution shifts induced by scene composition. To address this, we introduce a scene graph-based representation that focuses on task-relevant objects and relations, thereby mitigating sensitivity to irrelevant variation. Building on this idea, we develop a scene-graph skill learning framework that integrates graph neural networks with diffusion-based imitation learning, and further combine “focused” scene-graph skills with a vision-language model (VLM) based task planner. Experiments in both simulation and real-world manipulation tasks demonstrate substantially higher success rates than state-of-the-art baselines, highlighting improved robustness and compositional generalization in long-horizon tasks.
          </p>
        </div>
      </div>
    </div>
    <!-- Abstract. -->

  </div>
</section>

<section class="section">
  <div class="container is-max-desktop">
    <h2 class="title is-3">Overview</h2>
    <div class="content has-text-justified">
      <p>
        We propose a scene graph-based framework for skill composition in robot learning.
        Our method improves compositional generalization for long-horizon robotic manipulation
        by focusing on task-relevant object relations. The framework integrates graph neural networks,
        diffusion policy, and vision-language planning for robust real-world deployment.
      </p>
    </div>
  </div>
</section>
<!-- Overview. -->
<div class="content is-centered is-full-width has-text-centered">
  <img src="./static/images/mainfigure_v2.png" width="65%">
</div>
<!-- Overview. -->


<!-- Realworld Evaluation. -->
<section class="section">
  <div class="container is-max-desktop">
    <h2 class="title is-3">Real world experiment - Tool usage</h2>

    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="realworld-tool-usage" title="Real-world tool usage: scene graph-based skill composition" aria-label="Real-world tool usage experiment video" autoplay controls muted playsinline>
            <source src="./static/videos/tool_realworld.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Real-world robot manipulation: tool-usage task executed via scene graph-based atomic skills and skill composition for long-horizon generalization.
          </p>
        </div>
      </div>
    </div>
    <!-- Realworld Evaluation -->
  </div>
</section>
<!-- Realworld Evaluation. -->


<!-- Realworld Evaluation. -->
<section class="section">
  <div class="container is-max-desktop">
    <h2 class="title is-3">Real world experiment - Vegetable picking</h2>

    <!-- Realworld Evaluation -->
    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="realworld-vegetable-picking" title="Real-world vegetable picking: skill composition for manipulation" aria-label="Real-world vegetable picking experiment video" autoplay controls muted playsinline>
            <source src="./static/videos/veg_realworld.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Real-world robot vegetable picking demonstrating compositional generalization: scene graph representations focus on task-relevant objects and relations for robust execution.
          </p>
        </div>
      </div>
    </div>


    <!-- Realworld Evaluation -->
  </div>
</section>
<!-- Realworld Evaluation. -->


<!-- Simulation Evaluation. -->
<section class="section">
  <div class="container is-max-desktop">
    <h2 class="title is-3">Simulation experiments</h2>

    <!-- Simulation Evaluation -->
    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="sim-cube-out-and-in" title="Simulation: cube out-and-in task" aria-label="Simulation cube out-and-in task video" autoplay controls muted playsinline>
            <source src="./static/videos/sim1.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Simulation: cube out-and-in manipulation task showing scene graph-based atomic skills composed into a multi-step long-horizon policy.
          </p>
        </div>
      </div>
    </div>

    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="sim-sort-by-color" title="Simulation: sort by color" aria-label="Simulation sort-by-color task video" autoplay controls muted playsinline>
            <source src="./static/videos/sim2.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Simulation: sort-by-color task highlighting skill composition under distribution shift, with object-relation reasoning in a scene graph.
          </p>
        </div>
      </div>
    </div>

    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="sim-blocks-stacking" title="Simulation: blocks stacking game" aria-label="Simulation blocks stacking task video" autoplay controls muted playsinline>
            <source src="./static/videos/sim3.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Simulation: block stacking (long-horizon) demonstrating compositional robot learning with diffusion-based imitation learning and graph neural networks.
          </p>
        </div>
      </div>
    </div>

    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="sim-tools-usage" title="Simulation: tools usage" aria-label="Simulation tools usage task video" autoplay controls muted playsinline>
            <source src="./static/videos/sim4.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Simulation: tools-usage task illustrating scene graph-based skill learning and reliable execution when composing multiple atomic manipulation skills.
          </p>
        </div>
      </div>
    </div>

    <div class="columns">
      <div class="column is-centered">
        <div class="content has-text-centered">
          <video id="sim-obstacle-avoidance" title="Simulation: obstacle avoidance" aria-label="Simulation obstacle avoidance task video" autoplay controls muted playsinline>
            <source src="./static/videos/sim5.mp4"
                    type="video/mp4">
          </video>
          <p class="video-caption">
            Simulation: obstacle avoidance with focused scene graphs, improving robustness and compositional generalization for robotic manipulation.
          </p>
        </div>
      </div>
    </div>
    <!-- Simulation Evaluation -->
  </div>
</section>
<!-- Simulation Evaluation. -->


<footer class="footer">
  <div class="container">
    <!-- <div class="content has-text-centered">
      <a class="icon-link"
         href=""> -->
         <!-- ./static/videos/nerfies_paper.pdf -->
        <!-- <i class="fas fa-file-pdf"></i>
      </a> -->
      <!-- https://github.com/keunhong -->
      <!-- <a class="icon-link" href="" class="external-link" disabled>
        <i class="fab fa-github"></i>
      </a> -->
    <!-- </div> -->
    <div class="columns is-centered">
      <div class="column is-8">
        <div class="content">
          <p>
            This website is licensed under a <a rel="license"
                                                href="http://creativecommons.org/licenses/by-sa/4.0/">Creative
            Commons Attribution-ShareAlike 4.0 International License</a>.
          </p>
          <p>
            The source code of this website is borrowed from <a
              href="https://github.com/nerfies/nerfies.github.io">Nerfies</a>. We thank the authors for sharing the template.
          </p>
        </div>
      </div>
    </div>
  </div>
</footer>

</body>
</html>