forked from Sharma-Pragya/cadet-site
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
598 lines (559 loc) · 32 KB
/
Copy pathindex.html
File metadata and controls
598 lines (559 loc) · 32 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
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8">
<!-- 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="anonymized" content="https://anonymous.4open.science/w/cadet-site-C823/">
<meta name="description" content="<TODO> Paper on adaptive control placement for IoT closed-loop systems. MILCOM 2023. Authors: Pragya Sharma and Mani Srivastava. Evaluates local vs cloud control using a PyBullet-based vehicle following simulation.">
<meta name="keywords" content="<TODO> IoT, adaptive control, cloud computing, real-time systems, latency, MPC, PID, MILCOM 2023, PyBullet simulation, distributed control, Pragya Sharma, UCLA">
<meta property="og:title" content="SOCIAL MEDIA TITLE TAG"/>
<meta property="og:description" content="SOCIAL MEDIA DESCRIPTION TAG TAG"/>
<meta property="og:url" content="https://sharma-pragya.github.io/cadet-site/"/>
<!-- 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 dimensions 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>CADET: A Modular Platform for Evaluating Distributed Inference and Cooperative Autonomy in Connected Autonomous Vehicles</title>
<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>
</head>
<body>
<!-- Hero Section -->
<section class="hero is-large hero-white">
<div class="hero-body">
<div class="container">
<div class="columns is-centered">
<div class="column is-10 has-text-centered">
<h1 class="title is-1 hero-title-dark">CADET</h1>
<h2 class="subtitle is-3 hero-subtitle-dark">Cooperative Autonomy through Distributed Experimentation Toolkit</h2>
<div class="hero-tags">
<span class="tag is-primary is-medium">Modular</span>
<span class="tag is-primary is-medium">Reproducible</span>
<span class="tag is-primary is-medium">Hardware-in-the-Loop</span>
</div>
<div class="hero-meta">
<p class="is-size-5" style="color: #666;"><a href="https://sharma-pragya.github.io/" style="color: inherit; text-decoration: underline;">Pragya Sharma</a>, Brian Wang, Mani Srivastava • ICRA 2026</p>
</div>
<div class="hero-buttons">
<a href="https://drive.google.com/file/d/1_mv632DviNDyKGGgiP9gPIQR4qSsuWKe/preview" target="_blank" class="button is-primary is-medium is-rounded">
<span class="icon">
<i class="fas fa-video"></i>
</span>
<span>Watch Demo</span>
</a>
<button class="button is-primary is-medium is-rounded" disabled style="opacity: 0.5; cursor: not-allowed;">
<span class="icon">
<i class="fab fa-github"></i>
</span>
<span>Access Code</span>
</button>
<a href="ICRA_CADET.pdf" target="_blank" class="button is-primary is-medium is-rounded">
<span class="icon">
<i class="fas fa-file-pdf"></i>
</span>
<span>Read Paper</span>
</a>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Navigation Bar -->
<nav class="navbar is-fixed-top nav-sticky" role="navigation" aria-label="main navigation">
<div class="container">
<div class="navbar-brand">
<span class="navbar-item nav-brand">CADET</span>
</div>
<div class="navbar-menu">
<div class="navbar-end">
<a class="navbar-item" href="#motivation">Overview</a>
<a class="navbar-item" href="#comparison">Comparison</a>
<a class="navbar-item" href="#architecture">Architecture</a>
<a class="navbar-item" href="#netwaggle">NetWaggle</a>
<a class="navbar-item" href="#usecases">Evaluation</a>
<a class="navbar-item" href="#abstract">Abstract</a>
</div>
</div>
</div>
</nav>
<!-- Motivation Section -->
<section id="motivation" class="section motivation-section">
<div class="container">
<div class="columns is-multiline is-vcentered">
<div class="column is-12 has-text-centered">
<h2 class="title is-2 section-title">Motivation</h2>
<p class="subtitle is-5">Safety outcomes depend on the joint design of models, placement, and networks</p>
</div>
<div class="column is-12 has-text-centered">
<div class="publication-video coop-video" style="max-width: 800px; margin: 0 auto 2rem;">
<iframe src="https://drive.google.com/file/d/1zvf-io0BlmY-tmMEs0tVVSy0ASG4UNtJ/preview" allow="autoplay" allowfullscreen></iframe>
</div>
</div>
<div class="column is-4">
<div class="challenge-card">
<h3 class="title is-4">The Gap</h3>
<p>Existing AV platforms remain <strong>monolithic in deployment</strong>, executing entire pipelines on single onboard computers. V2X simulators model networks or modularity, but overlook cooperative autonomy as a <strong>distributed systems problem</strong> requiring compute platform heterogeneity and orchestration.</p>
</div>
</div>
<div class="column is-4">
<div class="challenge-card">
<h3 class="title is-4">CADET's Approach</h3>
<p>CADET <strong>disaggregates</strong> the AV stack into composable modules deployable across vehicles, RSUs, edge nodes, and cloud servers. NetWaggle provides <strong>reproducible network emulation</strong> with device heterogeneity, enabling systematic exploration of distributed execution strategies.</p>
</div>
</div>
<div class="column is-4">
<div class="challenge-card">
<h3 class="title is-4">What CADET Enables</h3>
<p>CADET allows systematic exploration of deployment topologies (V2V, V2I, cloud, edge) while isolating the impact of network characteristics (latency percentiles, jitter), compute resources (GPU architecture, multi-tenant contention), and model selection (accuracy-latency tradeoffs). Multi-level instrumentation traces causality from design choices to safety outcomes, revealing hidden failure modes invisible in single-tier evaluations.</p>
</div>
</div>
</div>
</div>
</section>
<!-- Platform Comparison -->
<section id="comparison" class="section comparison-section">
<div class="container">
<h2 class="title is-3 has-text-centered">Platform Comparison</h2>
<p class="subtitle is-5 has-text-centered">CADET integrates platform heterogeneity, orchestration, system-level metrics, and modularity</p>
<div class="table-container">
<table class="table is-bordered is-striped is-fullwidth comparison-table">
<thead>
<tr>
<th>Platform</th>
<th class="has-text-centered">V2X</th>
<th class="has-text-centered">Platform Heterogeneity</th>
<th class="has-text-centered">Network Realism</th>
<th class="has-text-centered">Orchestration</th>
<th class="has-text-centered">System-Level Metrics</th>
<th class="has-text-centered">Modularity</th>
</tr>
</thead>
<tbody>
<tr>
<td>Plexe</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
</tr>
<tr>
<td>Eclipse MOSAIC</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
</tr>
<tr>
<td>V2XVerse</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
</tr>
<tr>
<td>OpenCAMS</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
</tr>
<tr>
<td>OpenCDA</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
</tr>
<tr>
<td>VaN3Twin</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">○</td>
<td class="has-text-centered">●</td>
</tr>
<tr class="cadet-row">
<td><strong>CADET [ours]</strong></td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
<td class="has-text-centered">●</td>
</tr>
</tbody>
</table>
</div>
</div>
</section>
<!-- Paper abstract -->
<section id="abstract" class="section hero is-light">
<div class="container is-max-desktop">
<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>
Deep learning models are increasingly central to autonomous vehicle (AV) pipelines, yet their integration has traditionally followed a monolithic design where perception, planning, and control execute on a single onboard computer. This design overlooks the emerging paradigm of cooperative autonomy, where vehicles interact with roadside units (RSUs), edge servers, and cloud-hosted intelligence through vehicle-to-everything (V2X) connectivity. Cooperative perception and control improve safety and efficiency, but also introduce systems-level challenges: network latency, compute heterogeneity, and multi-tenant contention, all critically affect real-time decision-making. These challenges are further amplified by the increasing reliance on large foundation models, whose scale necessitates cloud deployment. We present CADET (Cooperative Autonomy through Distributed Experimentation Toolkit), a modular platform for systematic and reproducible evaluation of distributed cooperative autonomy systems under realistic deployment conditions. CADET decouples the AV stack into composable modules that can be flexibly deployed across vehicles, infrastructure, and edge/cloud tiers. The framework integrates state-of-the-art models, incorporates realistic network and workload emulation, and provides synchronized model-, system-, and task-level instrumentation. Through V2V and V2I experiments, we show that distributed deployment choices fundamentally shape safety, with V2V intent packets outperforming cloud-based perception and RSU-assisted perception sustaining safety until overloaded by concurrent requests. Although designed for AV pipelines, CADET also supports dataset-driven experimentation, enabling systems and ML researchers to benchmark distributed inference workloads independently of full vehicle simulation. CADET is open source, with code and demo available at <a href="https://anonymous.4open.science/w/cadet-site-C823/" target="_blank">https://anonymous.4open.science/w/cadet-site-C823/</a>.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- End paper abstract -->
<!-- Key Capabilities -->
<section class="section hero is-small">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-full">
<h2 class="title is-3">Platform Capabilities</h2>
<div class="content has-text-justified">
<div class="columns is-multiline">
<div class="column is-half">
<h3 class="title is-5">Modular Pipeline Architecture</h3>
<p>AV pipeline decomposed into Perception and Cognition stages with independently configurable sub-stages. Supports flexible deployment of state-of-the-art models (YOLO, MPC) and cooperation protocols across vehicle, RSU, edge, and cloud.</p>
</div>
<div class="column is-half">
<h3 class="title is-5">NetWaggle: Network Emulation</h3>
<p>Mininet-based network layer injects realistic V2X delay distributions, jitter patterns, packet loss rates, and bandwidth caps matching measured 4G-LTE, 5G-NR, DSRC, or Wi-Fi traces. Low overhead (<10ms) ensures repeatability.</p>
</div>
<div class="column is-half">
<h3 class="title is-5">Device Heterogeneity</h3>
<p>Client-server architecture supports deployment across embedded vehicle devices, GPU-equipped RSUs, and cloud servers. Asynchronous execution mode mimics real-world V2X timing behavior. Hardware-in-the-loop testing enabled.</p>
</div>
<div class="column is-half">
<h3 class="title is-5">Multi-Source Data Integration</h3>
<p>CARLA Simulator with Scenario Runner for configurable V2X scenarios. Dataset Mode for deterministic replay on Waymo, Argoverse, COCO, VQA2. Live Sensor support for physical hardware integration.</p>
</div>
<div class="column is-half">
<h3 class="title is-5">Orchestration and Batch Execution</h3>
<p>Single YAML configuration file specifies scenario parameters, model variants, deployment mappings, and network traces. SSH-based remote deployment. Systematic sweeps over vehicle density, delay profiles, or workloads.</p>
</div>
<div class="column is-half">
<h3 class="title is-5">Multi-Level Instrumentation</h3>
<p>Synchronized metrics across model-level (mAP, inference latency), system-level (CPU/GPU utilization, memory, energy), and task-level (time-to-collision, braking efficiency). Timestamped logs with NTP synchronization.</p>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- System Architecture -->
<section id="architecture" class="section hero is-light">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-full">
<h2 class="title is-3">CADET Architecture</h2>
<div class="content has-text-justified">
<p>
CADET implements a four-layer architecture that establishes clear abstraction boundaries while maintaining flexibility for diverse cooperative autonomy research scenarios.
</p>
<figure class="has-text-centered">
<img src="static/images/cadet-architecture.png" alt="CADET System Architecture" style="width:90%; height:auto; margin: 0 auto;">
<figcaption>CADET implements a modular four-layer architecture allowing components to be evaluated in isolation or integrated for end-to-end cooperative autonomy experiments.</figcaption>
</figure>
<p style="margin-top: 1.5rem;">
The <strong>Data Source Layer</strong> provides scenario generation via CARLA Simulator, dataset-driven evaluation, and live sensor integration. The <strong>Processing Core</strong> decomposes the AV pipeline into Perception and Cognition stages, each with independently configurable sub-stages deployable across the computational continuum. <strong>NetWaggle</strong> combines device heterogeneity with reproducible network emulation, supporting deployment from embedded vehicle-like devices to GPU-equipped RSUs and cloud servers. The <strong>Analysis Layer</strong> provides unified orchestration, synchronized multi-level instrumentation (model, system, task), and foundation model interfaces.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- NetWaggle Section -->
<section id="netwaggle" class="section hero is-small">
<div class="container is-max-desktop">
<div class="columns is-centered has-text-centered">
<div class="column is-full">
<h2 class="title is-3">NetWaggle: Standalone or Integrated</h2>
<div class="content">
<p class="subtitle is-6 has-text-centered" style="margin-bottom: 2rem;">
CADET components can be used <strong>together as an end-to-end platform</strong> or <strong>individually for targeted research</strong>. NetWaggle, CADET's network emulation layer, is available as a standalone tool for distributed systems research beyond autonomous vehicles.
</p>
<div class="columns is-vcentered">
<div class="column is-6">
<div class="box" style="height: 100%;">
<h3 class="title is-5">Standalone NetWaggle Use</h3>
<ul style="text-align: left; margin-left: 1rem;">
<li><strong>Network researchers</strong> can evaluate protocols under realistic V2X conditions without AV simulation</li>
<li><strong>Systems researchers</strong> can benchmark distributed inference across heterogeneous devices</li>
<li><strong>ML researchers</strong> can study model behavior under network-induced delays and jitter</li>
</ul>
<h3 class="title is-5" style="margin-top: 1.5rem;">Integrated CADET Use</h3>
<p style="text-align: left;">NetWaggle seamlessly integrates with CADET's Processing Core and Data Sources to enable end-to-end cooperative autonomy experiments with synchronized multi-level instrumentation.</p>
</div>
</div>
<div class="column is-6">
<figure>
<img src="static/images/netwaggle-deployment.png" alt="NetWaggle Architecture" style="width:100%; height:auto;">
<figcaption>NetWaggle network emulation with device heterogeneity across cloud, edge, and embedded platforms.</figcaption>
</figure>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Use Cases Section with Tabs -->
<section id="usecases" class="section usecases-section">
<div class="container">
<h2 class="title is-2 has-text-centered">Evaluation</h2>
<p class="subtitle is-5 has-text-centered">Distributed deployment choices fundamentally shape cooperative autonomy safety outcomes</p>
<div class="tabs is-centered is-boxed is-large">
<ul>
<li class="is-active" data-tab="v2v">
<a>
<span class="icon is-small"><i class="fas fa-car" aria-hidden="true"></i></span>
<span>V2V Cooperative Braking</span>
</a>
</li>
<li data-tab="v2i">
<a>
<span class="icon is-small"><i class="fas fa-broadcast-tower" aria-hidden="true"></i></span>
<span>V2I RSU-Assisted Perception</span>
</a>
</li>
<li data-tab="vlm">
<a>
<span class="icon is-small"><i class="fas fa-brain" aria-hidden="true"></i></span>
<span>Foundation Models</span>
</a>
</li>
</ul>
</div>
<div class="tab-content">
<!-- V2V Tab -->
<div class="tab-pane is-active" id="v2v">
<div class="columns is-vcentered">
<div class="column is-6">
<div class="publication-video">
<iframe src="https://drive.google.com/file/d/1d_2S4piI_ti3e-aHrH4sEdBB7zp11Mlt/preview" allow="autoplay" allowfullscreen></iframe>
</div>
</div>
<div class="column is-6">
<h3 class="title is-4">V2V Cooperative Braking</h3>
<p class="content">
Two-vehicle platoon scenario where the leader executes emergency braking and the follower must maintain safe inter-vehicle distance. Safety threshold set to 10m (approximately two car lengths). We compare V2V intent communication (DSRC, 20ms baseline, 50ms tail) against cloud-hosted YOLO11x perception (5G, 30ms baseline, 100ms tail).
</p>
<div class="result-highlight">
<p><strong>Result:</strong> V2V intent packets remain collision-free across all delay profiles. Cloud perception fails at p90 and p99 due to compounding delays (30-50ms inference + 100ms tail latency). Larger models impose higher energy costs (up to 1.66J per inference) without improving safety under adverse network conditions.</p>
</div>
<div class="columns">
<div class="column is-7">
<figure>
<img src="static/images/v2v-gap-analysis.png" alt="V2V Gap Analysis" style="width:100%; height:auto;">
<figcaption>Leader-follower gap over time at different speeds under perception-only and V2V-based braking. Safety violation occurs when gap ≤10m.</figcaption>
</figure>
</div>
<div class="column is-5">
<h4 class="title is-6">Collision Outcomes Across Network Delay Profiles</h4>
<div class="table-container">
<table class="table is-bordered is-narrow is-fullwidth collision-table">
<thead>
<tr>
<th>Policy</th>
<th>Model</th>
<th>Energy (J)</th>
<th>p50</th>
<th>p80</th>
<th>p90</th>
<th>p99</th>
</tr>
</thead>
<tbody>
<tr>
<td rowspan="1" class="v2v-row"><strong>V2V</strong></td>
<td>—</td>
<td>—</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
</tr>
<tr>
<td rowspan="5" class="perception-row"><strong>Perception<br>(Cloud)</strong></td>
<td>YOLO11n</td>
<td>0.73</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="unsafe-cell">✗</td>
<td class="unsafe-cell">✗</td>
</tr>
<tr>
<td>YOLO11s</td>
<td>0.76</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="unsafe-cell">✗</td>
<td class="unsafe-cell">✗</td>
</tr>
<tr>
<td>YOLO11m</td>
<td>0.92</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="unsafe-cell">✗</td>
<td class="unsafe-cell">✗</td>
</tr>
<tr>
<td>YOLO11l</td>
<td>1.27</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="unsafe-cell">✗</td>
<td class="unsafe-cell">✗</td>
</tr>
<tr>
<td>YOLO11x</td>
<td>1.66</td>
<td class="safe-cell">✓</td>
<td class="safe-cell">✓</td>
<td class="unsafe-cell">✗</td>
<td class="unsafe-cell">✗</td>
</tr>
</tbody>
</table>
</div>
<p class="is-size-7" style="margin-top: 0.5rem;"><strong>✓</strong> No Collision <strong>✗</strong> Collision</p>
</div>
</div>
</div>
</div>
</div>
<!-- V2I Tab -->
<div class="tab-pane" id="v2i">
<div class="columns is-vcentered">
<div class="column is-6">
<div class="publication-video">
<iframe src="https://drive.google.com/file/d/1NIHUo6EM4CGatJa2vK9nnVsABBZhGGNu/preview" allow="autoplay" allowfullscreen></iframe>
</div>
</div>
<div class="column is-6">
<h3 class="title is-4">V2I-Assisted Braking</h3>
<p class="content">
Ego vehicle travels at 20mph when a pedestrian emerges from behind an occluding truck. Safety margin defined as the difference between time-to-collision (TTC) and time-to-event (TTE). We evaluate four configurations: onboard perception, cloud-hosted perception, RSU-local V2I, and RSU-cloud V2I under varying visibility conditions and concurrent client loads.
</p>
<div class="result-highlight">
<p><strong>Result:</strong> Perception-only policies achieve sufficient safety margins under clear visibility but deteriorate rapidly as conditions worsen. RSU-local and RSU-cloud V2I maintain positive margins across all visibility settings. RSU inference latency exceeds 100ms with 5-10 extra clients. Cloud p99 latencies exceed 200ms with 10 clients due to queuing under concurrent load.</p>
</div>
<div class="columns">
<div class="column is-6">
<figure>
<img src="static/images/latency-violin-plot.png" alt="V2I Latency" style="width:100%; height:auto;">
<figcaption>End-to-end latency distribution under increasing concurrent load (0, 5, 10 extra vehicles) for PC and RSU deployments.</figcaption>
</figure>
</div>
<div class="column is-6">
<figure>
<img src="static/images/v2i-safety-boxplot.png" alt="Safety Analysis" style="width:100%; height:auto;">
<figcaption>Safety margin (TTC-TTE) across visibility conditions. Positive values indicate safe operation.</figcaption>
</figure>
</div>
</div>
</div>
</div>
</div>
<!-- VLM Tab -->
<div class="tab-pane" id="vlm">
<div class="columns is-centered">
<div class="column is-10">
<h3 class="title is-4">Dataset Mode with Foundation Models</h3>
<p class="content">
CADET's dataset mode enables evaluation of vision-language models independently of full vehicle simulation. We evaluate ten VLMs across three GPU servers (NVIDIA A6000, A100, A16) using the VQA2 dataset as a proxy for cooperative autonomy scenarios where sensor data is queried in natural language.
</p>
<div class="result-highlight">
<p><strong>Result:</strong> Inference latencies exceed 200ms even for smaller models, far beyond the deadlines of safety-critical coordination. 13B-scale models require >25GB GPU memory and deliver throughput below 10 tokens/s, while 3B-7B models sustain >12 tokens/s with manageable memory footprints. Although semantic queries are promising for higher-level coordination, current models are too slow and resource-intensive for low-level safety decisions.</p>
</div>
<figure class="has-text-centered">
<img src="static/images/accuracy-latency-plot.png" alt="VLM Accuracy-Latency" style="width:80%; height:auto; margin: 0 auto;">
<figcaption>Accuracy-latency tradeoffs for VLMs across heterogeneous GPU servers (A6000, A100, A16). Models and devices exhibit significant variation in the accuracy-latency tradeoff space.</figcaption>
</figure>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- Paper poster -->
<!-- <section class="hero is-small is-light">
<div class="hero-body">
<div class="container">
<h2 class="title">Poster</h2>
<img src="static/pdfs/poster.png" alt="Poster", style="max-width:100%; height:auto;">
</img>
</div>
</div>
</section> -->
<!--End paper poster -->
<!--BibTex citation -->
<!-- section class="section" id="BibTeX">
<div class="container is-max-desktop content">
<h2 class="title">BibTeX</h2>
<pre><code>@inproceedings{sharma2023impact,
title={Impact of Delays and Computation Placement on Sense-Act Application Performance in IoT},
author={Sharma, Pragya and Srivastava, Mani B},
booktitle={MILCOM 2023-2023 IEEE Military Communications Conference (MILCOM)},
pages={133--138},
year={2023},
organization={IEEE}}
</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>