<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>arXiv Daily Briefing — cs.CV · cs.RO</title>
    <link>https://gisbi-kim.github.io/arxiv-daily-summary/</link>
    <atom:link href="https://gisbi-kim.github.io/arxiv-daily-summary/feed.xml" rel="self" type="application/rss+xml" />
    <description>cs.CV · cs.RO arXiv new submissions, bucketed by lab ROI (3D/Scene, Robot Learning, Autonomous Driving, Foundation Models, Generation, Efficiency, Embodied AI, Safety). Daily briefing with insights and cross-domain contrast.</description>
    <language>ko</language>
    <lastBuildDate>Wed, 06 May 2026 15:15:33 +0900</lastBuildDate>
    <ttl>720</ttl>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-05-06 (수)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-06.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-06.html</guid>
      <pubDate>Wed, 06 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 분명한 흐름은 "World Model 평가가 'reconstruction loss → reward alignment + interactive bench'로 paradigm 전환" 한 거예요. 어제까지가 VLA의 'monitoring·safety·test-time compute' 운영 layer 표면화였다면, 오늘은 그 substrate인 World Model 측에서 한 번에 두 결이 나란히 등장 — RoboAlign-R1 이 robot video WM을 'reconstruction loss + perceptual similarity' 같은 low-level objective 대신 'instruction following · manipulation succes…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-05-05 (화)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-05.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-05.html</guid>
      <pubDate>Tue, 05 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 분명한 흐름은 VLA가 'monitoring·safety·test-time compute'라는 새 layer로 급격히 확장 한 것이에요. 어제까지가 "pixel-free latent world-action"이라는 substrate 측 paradigm shift였다면, 오늘은 그 위에 얹는 운영 layer가 한 번에 표면화한 자리 — Sentinel-VLA (metacognitive monitoring + on-demand reasoning) + VLA-ATTC (adaptive test-time compute + Relative Action Critic) + TAIL-Safe (IL safety set via Lipschitz Q) + To Do or Not…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-05-04 (월)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-04.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-04.html</guid>
      <pubDate>Mon, 04 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 단단한 흐름은 VLA latent reasoning paradigm이 2주차에 들어선 것 입니다. 지난주 LaST-R1·MotuBrain·PRTS가 "latent CoT + RL" 정조준한 자리에서, 이번주는 Being-H0.7 (latent world-action without future frame) + Hamiltonian World Models (physics-grounded latent dynamics) + IVLR (95.5% long-horizon)이 나란히 등장. 한 주만에 같은 paradigm이 다른 axis로 누적되는 건 burst가 아니라 community standard 정착 의 분명한 신호고, 동시에 World Model Survey…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-05-03 (일)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-03.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-03.html</guid>
      <pubDate>Sun, 03 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[주말이 지난 일요일 시점에서 같은 금요일 배치를 다시 보면, 3D/Scene과 Robot Learning이 deployment 라인의 양 축으로 자리잡은 게 더 분명하게 보여요 . 7일 전(2026-04-26 동급 스냅샷) 대비 3D/Scene이 20편 으로 굳었고(7일 전 13편 → +54%), Robot Learning도 17편으로(+13%) 한 주 안에서 단단해진 자리. pastweek 598편(CV) + 187편(RO) 누적 안에서도 두 라인이 가장 dense한 자리고, 특히 sparse-view 3DGS 4편이 한 날 등장한 게 라인 성숙의 측면에서 가장 분명한 신호입니다. 한 주 더 지나면 ScanNet++/Mip-NeRF 360 측 새 SOTA 보고가 누적될…]]></description>
    </item>
    <item>
      <title>🗓 arXiv Weekly Retrospective</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-02-weekly.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-02-weekly.html</guid>
      <pubDate>Sat, 02 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[arXiv Daily Briefing 2026-05-02]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-05-01 (금)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-01.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-05-01.html</guid>
      <pubDate>Fri, 01 May 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 두드러진 매크로 신호는 3D/Scene이 다시 살아났다 는 점이에요. 7일 전 13편 → 오늘 20편으로 +54% 급등했고, 그 안에서 sparse-view 3DGS 결만 4편( RecGen · Sparse-View 3DGS in the Wild · Residual GS for CBCT · Generalizable Sparse-View Recon ) 동시에 등장했습니다. 어제까지 '3DGS deployment의 production cost'였다면 오늘은 'sparse view에서도 robust한가'로 community focus가 이동한 모양새고, 이건 driving simulation·embodied data scaling 모두에 직접 영향을 끼치는 자리라…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-30 (목)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-30.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-30.html</guid>
      <pubDate>Thu, 30 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 두드러진 매크로 신호는 Generation 76편 으로 한 주 내내 1위를 잡고 있다는 점이에요. 어제 66편 → 오늘 76편으로 +15% 더 붙었고, 오늘 /new 15편으로도 상위 버킷이라 '생성 모델 라인의 광폭 누적'이 6일째 끊기지 않고 있습니다. 의료 video generation·SR·diffusion sampling acceleration·post-train 측 결들이 한꺼번에 누적되는 모양새고, 이는 어제까지 지적했던 '단독 신규 모델보다 deployment·post-train'으로 무게가 옮겨가는 흐름의 정점이라고 봐요. 반면 어제 절정이었던 Safety/Alignment 는 67편 → 26편으로 -61% 급락했는데, 이건 분류 keywor…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-29 (수)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-29.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-29.html</guid>
      <pubDate>Wed, 29 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 가장 강한 매크로 신호는 Safety/Alignment 가 31→67편으로 +116% 나 폭증했다는 점이에요. 솔직히 한 주 사이에 두 배가 넘게 뛰는 건 흔치 않은 일이고, 어제까지 Safety가 39편이었던 것에서도 +72%가 더 붙은 거라 우연이 아닙니다. 의료 영상 robustness audit, foundation segmentation의 도메인 shift, deepfake detection robustness, label noise·OOD·adversarial 같은 정통 라인이 한꺼번에 쏟아지는 모양새고, 어제까지 보던 'production-deployment 라인이 굵어진다'는 흐름의 정점이 이번주에 찍혔다고 봐요. 반면 Foundation Model…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-28 (화)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-28.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-28.html</guid>
      <pubDate>Tue, 28 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 pastweek 누적의 가장 큰 신호는 Autonomous Driving 이 13편 → 32편으로 +146% 폭증했다는 점이에요. 어제까지 'AD가 perception 단독 결을 줄이고 큰 모델 안으로 흡수된다'고 보던 흐름이 이번주 다시 반전 — V2X coordination(SwarmDrive), 4D occupancy directing(OccDirector, 어제 결), end-to-end parking(ParkingScenes), pedestrian intention prediction(ESIA), trajectory planning + LLM negotiation 같은 결들이 동시에 쌓이면서 AD가 '단순 perception'에서 'social-aware…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-27 (월)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-27.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-27.html</guid>
      <pubDate>Mon, 27 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 pastweek 누적을 살펴보면, 가장 큰 반전은 Foundation Models 가 69편(지난주) → 111편으로 +61% 다시 폭증한 점이에요. 어제까지 'VLM이 호흡 고른다'고 봤는데 한 주 단위로 보니 호흡이라기보단 주제가 갈아탔다 가 맞습니다 — chart-to-code, multi-image grounding, surgical safety reasoning, scene graph hallucination 같이 'VLM의 어디가 깨지나'를 정조준한 신규 페이퍼들이 동시에 쏟아져요. 한편 Robot Learning 도 48 → 71편( +48% )로 VLA 라인이 다시 가속하고, 3D/Scene 은 34 → 46편( +35% )로 SLAM·dynamic…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-26 (일)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-26.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-26.html</guid>
      <pubDate>Sun, 26 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 pastweek 누적을 살펴보면, 가장 큰 변화는 Safety/Alignment 가 31편(지난주) → 58편으로 거의 두 배 가까이 부풀어 오른 거예요. CV쪽 TTA·OOD·adversarial·face forgery 라인이 동시에 폭발하면서 일어난 현상입니다. 한편 Foundation Models 는 112편 → 69편으로 -38% 빠졌는데, 일주일 단위로 보면 'VLM 폭주'가 잠시 호흡을 고르고 'VLM이 어디서 틀리는가'로 무게추가 옮겨가는 신호라고 봅니다. Generation은 +20%, Embodied AI는 +33%, Efficiency는 +21% 로 완만한 상승이라 dynamic이 살아있어요.]]></description>
    </item>
    <item>
      <title>🗓 arXiv Weekly Retrospective</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-25-weekly.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-25-weekly.html</guid>
      <pubDate>Sat, 25 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[arXiv Daily Briefing 2026-04-25]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-24 (금)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-24.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-24.html</guid>
      <pubDate>Fri, 24 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 pastweek 누적을 훑어보니 제일 뜨거운 축은 여전히 Foundation Models 하고 Generation 인데 (어제 스냅샷 각각 CV 107·75편), 오늘 /new만 떼어 보면 결이 꽤 다릅니다. VLM 쪽이 "더 큰 모델 + 더 큰 벤치"에서 이제 "왜 틀리는지 진단"으로 무게추가 넘어오는 중이에요. DistortBench 류의 low-level 진단, Seeing Isn't Believing(평가자 VLM 맹점), Prompt-Induced Hallucinations, Symbolic Grounding 이 하루에 같이 나왔거든요. 추론 능력이 아니라 재현율·신뢰성 이 다음 경쟁 축이 되는 분기점으로 보입니다.]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-23 (목)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-23.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-23.html</guid>
      <pubDate>Thu, 23 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번주 pastweek을 훑어보니 제일 두드러진 건 Foundation Models 버킷입니다. CV 쪽 107편이 몰렸고, 그중 VLM hallucination·token pruning·visual reasoning 쪽이 특히 두꺼워요. 오늘만 해도 LLM-as-Judge로 tone-induced hallucination을 측정하는 논문, zero-cost contrastive editing, STEPSTEM 같은 fine-grained MLLM 평가 벤치마크가 한꺼번에 쏟아졌네요. Generation 도 79편으로 두번째인데 "world model 예측"과 "robot policy"를 엮는 방향이 본격적으로 올라오고 있어요. 오늘의 Mask World Model과 R…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-22 (수)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-22.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-22.html</guid>
      <pubDate>Wed, 22 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번 주 가장 뜨거운 버킷은 압도적으로 Robot Learning 입니다. cs.RO 전체 109편 중 manipulation 관련만 12편, vision-language 키워드가 9편, grasping 8편이니까 거의 3분의 1이 로봇 학습 관련이에요. 특히 VLA(Vision-Language-Action) 모델이 이번 주에만 10편 이상 쏟아졌는데, 오늘 나온 π₀.₇ (Physical Intelligence)이 결정적입니다. 단순 language command를 넘어 메타데이터, 서브골 이미지, 심지어 실패 데이터까지 conditioning으로 활용해서 "steerable foundation model"을 만들겠다는 건데, 이건 VLA 패러다임을 한 단계 끌어올리는…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-21 (화)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-21.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-21.html</guid>
      <pubDate>Tue, 21 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번 주 가장 뜨거운 버킷은 또다시 Robot Learning 입니다. 지난주에 이어 VLA(Vision-Language-Action) 모델의 진화가 가속되고 있는데, 오늘 나온 π₀.₇ 이 단연 압권이에요. Physical Intelligence 팀이 만든 이 모델은 언어 조건부 다단계 태스크, 제로샷 크로스-엠바디먼트 일반화, 심지어 에스프레소 머신 조작까지 해내면서 "조종 가능한 범용 로봇 FM"의 가능성을 보여줬습니다. 이와 함께 VLA에 장기 기억을 붙이는 연구(Long-Term Memory for VLA), VLA를 fine-tuning할 때 VQA 능력이 붕괴하는 문제를 잡는 AEGIS, 시뮬레이터 없이 조작 데이터를 생성하는 ShapeGen 등이 한꺼번에…]]></description>
    </item>
    <item>
      <title>📄 arXiv Daily Briefing — 2026-04-18 (금)</title>
      <link>https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-18.html</link>
      <guid isPermaLink="true">https://gisbi-kim.github.io/arxiv-daily-summary/posts/2026-04-18.html</guid>
      <pubDate>Sat, 18 Apr 2026 09:00:00 +0900</pubDate>
      <description><![CDATA[이번 주 제일 뜨거운 버킷은 단연 Robot Learning 이에요. cs.RO 전체 116편 중 조작·학습 관련 논문이 28편으로 거의 4분의 1을 차지했는데, 특히 VLA(Vision-Language-Action) 모델이 단순 action regression을 넘어 world model과 value function을 내장하는 방향으로 진화하는 흐름이 뚜렷합니다. 오늘만 해도 WAV model, HiVLA, R3D 같은 구조적으로 한 단계 올라간 VLA 논문이 한꺼번에 나왔어요. 3D/Scene 버킷도 여전히 강세인데, 3DGS(3D Gaussian Splatting)가 이제 순수 렌더링을 완전히 벗어나서 SLAM, 시뮬레이션, 심지어 4D 레이더까지 침투하고 있습니다…]]></description>
    </item>
  </channel>
</rss>
