Skip to content

Knowledge Graph

Jump to bottom
hwoo.han edited this page Jul 25, 2026 · 2 revisions

πŸ•ΈοΈ Knowledge Graph β€” navigate the wiki visually

A visual map of how the wiki's pages connect. Click any node to jump to that page (GitHub renders these Mermaid graphs; if your viewer disables click-through, use the wikilink legend under each graph). Five views: the overall map, how the cross-paper reviews interconnect, the model lineages, the world-model thematic cluster, and the RSS 2026 improvement-loop cluster.

← Back to Home Β· full list in the sidebar.

1. Overall map β€” three ways in

flowchart TB
  HOME(["🏠 Home"])
  HOME --> V["By venue"]
  HOME --> R["By topic review"]
  HOME --> L["By model lineage"]
  HOME --> F["Foundational refs"]

  V --> RSS["RSS 2026 πŸ†•"]
  V --> ICLR["ICLR 2026"]
  V --> ICRA["ICRA 2026"]
  V --> ICML["ICML 2026"]
  V --> CVPR["CVPR 2026/25"]
  V --> NEUR["NeurIPS 2025"]
  V --> CORL["CoRL 2025"]
  V --> IROS["IROS 2025"]

  R --> ARCHm["VLA Architectures<br/>(hub review)"]
  R --> WMm["World Models"]
  R --> RLm["RL for VLA"]
  R --> more["+ 8 more reviews β†’<br/>see graph 2"]

  L --> pim["Ο€-series"]
  L --> grootm["GR00T N1β†’N1.7"]
  L --> rtcm["RTC β†’ REMAC"]

  click RSS "https://github.com/Heungwoo/research/wiki/RSS-2026-VLA-Manipulation-Survey"
  click ICML "https://github.com/Heungwoo/research/wiki/ICML-2026"
  click ICLR "https://github.com/Heungwoo/research/wiki/ICLR-2026-VLA-Manipulation-Survey"
  click ICRA "https://github.com/Heungwoo/research/wiki/ICRA-2026-VLA-Manipulation-Survey"
  click CVPR "https://github.com/Heungwoo/research/wiki/CVPR-2026-VLA-Manipulation-Survey"
  click NEUR "https://github.com/Heungwoo/research/wiki/NeurIPS-2025-VLA-Manipulation-Survey"
  click CORL "https://github.com/Heungwoo/research/wiki/CoRL-2025-VLA-Manipulation-Survey"
  click IROS "https://github.com/Heungwoo/research/wiki/IROS-2025-VLA-Manipulation-Survey"
  click ARCHm "https://github.com/Heungwoo/research/wiki/Review-VLA-Architecture"
  click WMm "https://github.com/Heungwoo/research/wiki/Review-World-Models"
  click RLm "https://github.com/Heungwoo/research/wiki/RL"
Loading

Legend: Venues β†’ RSS 2026 πŸ†• Β· ICLR 2026 Β· ICRA 2026 Β· ICML 2026 Β· CVPR 2026 Β· NeurIPS 2025 Β· CoRL 2025 Β· IROS 2025.

2. Cross-paper reviews β€” how the topics interconnect

flowchart LR
  ARCH["VLA Architectures"]
  WM["World Models"]
  VLMA["VLM↔Action"]
  ATTN["VLA Attention"]
  SYS["System 0/1/2"]
  RL["RL for VLA"]
  MEM["VLA Memory"]
  XEMB["Cross-Embodiment"]
  GOAL["Goal-Image Cond."]
  DEX["Dexterous Manip."]
  WAM["WAM vs VLA Robustness"]
  MLF["ML foundations"]

  TAC["Tactile VLA"]
  HUM["Humanoid VLA"]

  ARCH ---|"Category E, expanded"| WM
  ARCH ---|"wiring axis"| VLMA
  ARCH ---|"attention axis"| ATTN
  ARCH --- XEMB
  ARCH --- MEM
  ARCH --- DEX
  WM ---|"empirical check"| WAM
  WM ---|"WM-as-prompt overlap"| GOAL
  WM ---|"world-model RFT"| RL
  GOAL --- ARCH
  SYS ---|"S2β†’S1 hand-off"| VLMA
  SYS --- DEX
  DEX --- XEMB
  MEM --- XEMB
  MLF --- ATTN
  DEX ---|"predict future contact<br/>(RSS 2026: ViTacFormer, CGP)"| TAC
  TAC --- ARCH
  HUM --- SYS
  HUM ---|"Ξ¨β‚€, HoMMI (RSS 2026)"| XEMB
  RL ---|"improvement loop<br/>(RSS 2026: RECAP et al.)"| ARCH

  click TAC "https://github.com/Heungwoo/research/wiki/Review-Tactile-VLA"
  click HUM "https://github.com/Heungwoo/research/wiki/Review-Humanoid-VLA"

  click ARCH "https://github.com/Heungwoo/research/wiki/Review-VLA-Architecture"
  click WM "https://github.com/Heungwoo/research/wiki/Review-World-Models"
  click VLMA "https://github.com/Heungwoo/research/wiki/Review-VLM-Action-Connection"
  click ATTN "https://github.com/Heungwoo/research/wiki/Review-VLA-Attention"
  click SYS "https://github.com/Heungwoo/research/wiki/Review-System-0-1-2"
  click RL "https://github.com/Heungwoo/research/wiki/RL"
  click MEM "https://github.com/Heungwoo/research/wiki/Review-VLA-Memory"
  click XEMB "https://github.com/Heungwoo/research/wiki/Review-Cross-Embodiment"
  click GOAL "https://github.com/Heungwoo/research/wiki/Review-Goal-Image-Conditioning"
  click DEX "https://github.com/Heungwoo/research/wiki/Review-Dexterous-Manipulation"
  click WAM "https://github.com/Heungwoo/research/wiki/Review-WAM-vs-VLA-Robustness"
  click MLF "https://github.com/Heungwoo/research/wiki/ML"
Loading

Legend: VLA Architectures Β· World Models Β· VLM↔Action Β· VLA Attention Β· System 0/1/2 Β· RL for VLA Β· VLA Memory Β· Cross-Embodiment Β· Goal-Image Conditioning Β· Dexterous Manipulation Β· Tactile VLA πŸ†• Β· Humanoid VLA πŸ†• Β· WAM vs VLA Robustness Β· ML foundations.

3. Model lineages

flowchart LR
  subgraph PI["Ο€-series (Physical Intelligence)"]
    p0["Ο€0"] --> p05["Ο€0.5"] --> p06["Ο€0.6"] --> rec["Ο€*0.6 / RECAP<br/>(RSS 2026 oral)"] --> p07["Ο€0.7"]
  end
  subgraph GR["NVIDIA GR00T"]
    g1["N1"] --> g15["N1.5"] --> g16["N1.6"] --> g17["N1.7"]
  end
  subgraph OV["OpenVLA family"]
    ov["OpenVLA"] --> oft["OpenVLA-OFT / FASTER"]
  end
  subgraph RTC["Real-time chunking line"]
    rtc["RTC"] --> ttr["training-time RTC<br/>(PI, ext.)"] --> rem["REMAC"] --> asy["AsyncVLA"]
    rtc -->|"internalized"| leg["Legato (RSS 2026)"]
    ttr -.->|"adopted by"| psi["Ξ¨β‚€ deployment"]
  end
  subgraph LBM["TRI LBM co-training line πŸ†•"]
    lbm1["ICLR study"] --> lbm2["89-policy study<br/>(RSS 2026)"]
  end
  subgraph LIB["LIBERO robustness line πŸ†•"]
    lib0["LIBERO"] --> libp["LIBERO-Plus"] --> libx["LIBERO-X<br/>(RSS 2026)"]
  end

  click leg "https://github.com/Heungwoo/research/wiki/RSS-2026-Legato"
  click psi "https://github.com/Heungwoo/research/wiki/Review-Psi0"
  click lbm1 "https://github.com/Heungwoo/research/wiki/Review-LBM-Cotraining"
  click lbm2 "https://github.com/Heungwoo/research/wiki/RSS-2026-LBM-Cotraining-Study"
  click libp "https://github.com/Heungwoo/research/wiki/CVPR-2026-LIBERO-Plus"
  click libx "https://github.com/Heungwoo/research/wiki/RSS-2026-LIBERO-X"
  click p05 "https://github.com/Heungwoo/research/wiki/CoRL-2025-pi05"
  click p06 "https://github.com/Heungwoo/research/wiki/PI-pi06"
  click rec "https://github.com/Heungwoo/research/wiki/PI-RECAP"
  click p07 "https://github.com/Heungwoo/research/wiki/PI-pi07"
  click g17 "https://github.com/Heungwoo/research/wiki/Review-GR00T-Series"
  click g1 "https://github.com/Heungwoo/research/wiki/Review-GR00T-Series"
  click ov "https://github.com/Heungwoo/research/wiki/CoRL-2024-OpenVLA"
  click oft "https://github.com/Heungwoo/research/wiki/ICLR-2026-FASTER"
  click rtc "https://github.com/Heungwoo/research/wiki/NeurIPS-2025-Real-Time-Chunking"
  click rem "https://github.com/Heungwoo/research/wiki/ICLR-2026-Masked-Action-Chunking"
  click asy "https://github.com/Heungwoo/research/wiki/Review-AsyncVLA"
Loading

Legend: Ο€ β€” Ο€ series evolution Β· Ο€0.5 Β· Ο€0.6 Β· Ο€*0.6/RECAP Β· Ο€0.7 (long-form). GR00T β€” N1β†’N1.7. OpenVLA β€” OpenVLA β†’ FASTER. RTC β€” RTC β†’ REMAC β†’ AsyncVLA; RSS 2026 branch: Legato (native continuation) and training-time RTC adopted in Ξ¨β‚€. LBM β€” ICLR study β†’ RSS 89-policy study. LIBERO β€” LIBERO-Plus β†’ LIBERO-X.

4. World-model cluster (what predicts Γ— how used)

flowchart TB
  WMR["World Models review"]
  WMR --> back["Backbone / VAM"]
  WMR --> uwm["Unified WM + policy πŸ†•"]
  WMR --> rlenv["RL environment"]
  WMR --> data["Data factory"]
  WMR --> plan["Planner / IDM-decode"]

  back --> cosmos["Cosmos-Policy"]
  back --> genie["Genie-Envisioner"]
  back --> mimic["mimic-video<br/>(RSS 2026, 10Γ— sample-eff.)"]
  uwm --> lda["LDA-1B<br/>(RSS 2026, DINO-latent dynamics)"]
  rlenv --> wmpo["WMPO"]
  rlenv --> ctrl["Ctrl-World"]
  rlenv --> wgym["WorldGym"]
  data --> dgen["DreamGen"]
  data --> qrw["Qwen-RobotWorld<br/>(language-actioned)"]
  plan --> goalv["Goal-VLA"]
  WMR --> aux["Aux loss β†’ DreamVLA"]
  WMR ---|"empirical check"| wamr["WAM vs VLA Robustness"]
  WMR ---|"WM-as-prompt"| goalc["Goal-Image review"]

  click mimic "https://github.com/Heungwoo/research/wiki/RSS-2026-mimic-video"
  click lda "https://github.com/Heungwoo/research/wiki/RSS-2026-LDA-1B"
  click qrw "https://github.com/Heungwoo/research/wiki/Review-Qwen-RobotWorld"
  click WMR "https://github.com/Heungwoo/research/wiki/Review-World-Models"
  click cosmos "https://github.com/Heungwoo/research/wiki/ICLR-2026-Cosmos-Policy"
  click genie "https://github.com/Heungwoo/research/wiki/ICLR-2026-Genie-Envisioner"
  click wmpo "https://github.com/Heungwoo/research/wiki/ICLR-2026-WMPO"
  click ctrl "https://github.com/Heungwoo/research/wiki/ICLR-2026-Ctrl-World"
  click wgym "https://github.com/Heungwoo/research/wiki/ICLR-2026-WorldGym"
  click dgen "https://github.com/Heungwoo/research/wiki/CoRL-2025-DreamGen"
  click goalv "https://github.com/Heungwoo/research/wiki/ICRA-2026-Goal-VLA"
  click aux "https://github.com/Heungwoo/research/wiki/NeurIPS-2025-DreamVLA"
  click wamr "https://github.com/Heungwoo/research/wiki/Review-WAM-vs-VLA-Robustness"
  click goalc "https://github.com/Heungwoo/research/wiki/Review-Goal-Image-Conditioning"
Loading

Legend: World Models Β· Cosmos-Policy Β· Genie-Envisioner Β· mimic-video πŸ†• Β· LDA-1B πŸ†• Β· WMPO Β· Ctrl-World Β· WorldGym Β· DreamGen Β· Qwen-RobotWorld πŸ†• Β· Goal-VLA Β· DreamVLA Β· WAM vs VLA Robustness Β· Goal-Image Conditioning.

5. RSS 2026 cluster β€” the improvement loop πŸ†•

flowchart TB
  RSSV["RSS 2026 survey"]
  RSSV --> T1["1 Β· RL from experience"]
  RSSV --> T2["2 Β· Human-video transfer"]
  RSSV --> T3["3 Β· Video/world models"]
  RSSV --> T4["4 Β· Contact as representation"]
  RSSV --> T5["5 Β· Cross-embodiment hands"]
  RSSV --> T6["6 Β· Evaluation infrastructure"]

  T1 --> recap["Ο€*0.6 / RECAP"]
  T2 --> h2r["H2R Emergence (PI)<br/>co-train, emerges w/ diversity"]
  T2 --> psi0["Ψ₀ (in-depth)<br/>decouple: video→VLM, robot→expert"]
  T2 --> hommi["HoMMI"]
  h2r ---|"⇄ the open fork"| psi0
  T3 --> mimicv["mimic-video (VAM)"]
  T3 --> lda1b["LDA-1B (unified WM)"]
  T4 --> vitac["ViTacFormer"]
  T4 --> cgp["Contact-Grounded Policy"]
  T5 --> dgz["DexGrasp-Zero"]
  T5 --> ohra["One Hand to Rule Them All"]
  T6 --> polaris["PolaRiS (real-to-sim)"]
  T6 --> libx2["LIBERO-X"]
  RSSV --> lbms["LBM co-training study (TRI)<br/>cross-cutting data anchor"]
  RSSV --> mech["Inference mechanics:<br/>OAT Β· Legato"]

  click RSSV "https://github.com/Heungwoo/research/wiki/RSS-2026-VLA-Manipulation-Survey"
  click recap "https://github.com/Heungwoo/research/wiki/PI-RECAP"
  click h2r "https://github.com/Heungwoo/research/wiki/RSS-2026-Human2Robot-Emergence"
  click psi0 "https://github.com/Heungwoo/research/wiki/Review-Psi0"
  click hommi "https://github.com/Heungwoo/research/wiki/RSS-2026-HoMMI"
  click mimicv "https://github.com/Heungwoo/research/wiki/RSS-2026-mimic-video"
  click lda1b "https://github.com/Heungwoo/research/wiki/RSS-2026-LDA-1B"
  click vitac "https://github.com/Heungwoo/research/wiki/RSS-2026-ViTacFormer"
  click cgp "https://github.com/Heungwoo/research/wiki/RSS-2026-Contact-Grounded-Policy"
  click dgz "https://github.com/Heungwoo/research/wiki/RSS-2026-DexGrasp-Zero"
  click ohra "https://github.com/Heungwoo/research/wiki/RSS-2026-One-Hand"
  click polaris "https://github.com/Heungwoo/research/wiki/RSS-2026-PolaRiS"
  click libx2 "https://github.com/Heungwoo/research/wiki/RSS-2026-LIBERO-X"
  click lbms "https://github.com/Heungwoo/research/wiki/RSS-2026-LBM-Cotraining-Study"
  click mech "https://github.com/Heungwoo/research/wiki/RSS-2026-OAT"
Loading

Legend: RSS 2026 survey Β· Ο€*0.6+RECAP Β· H2R Emergence Β· Ξ¨β‚€ Β· HoMMI Β· mimic-video Β· LDA-1B Β· ViTacFormer Β· Contact-Grounded Policy Β· DexGrasp-Zero Β· One-Hand Β· PolaRiS Β· LIBERO-X Β· LBM co-training study Β· OAT Β· Legato.

Note on interactivity: GitHub renders Mermaid in wikis and supports node click-through to URLs. A fully interactive force-directed graph (Obsidian/Foam-style) would need JavaScript, which GitHub's wiki sandbox strips β€” so this page uses Mermaid (static but clickable) plus the wikilink legends. If you clone/serve the wiki in Obsidian or Foam, the wiki's internal links across all pages already form a live force-directed graph there.

← Back to Home

Navigation

πŸ“– Reviews

🏷 Model lineages

🧠 ML foundations

πŸ—“ Conferences

(each page indexes its per-paper pages)

πŸ“Œ Foundational

Clone this wiki locally