[
  {
    "name": "UniPi",
    "org": null,
    "year": 2023,
    "paradigm": "inverse dynamics",
    "backbone": "Video diffusion (~1.7B)",
    "notes": "Early recipe: generate the future video, then infer the actions that produce it.",
    "url": "https://proceedings.neurips.cc/paper_files/paper/2023/file/1d5b9233ad716a43be5c0d3023cb82d0-Paper-Conference.pdf",
    "slug": "unipi"
  },
  {
    "name": "GR-1",
    "org": null,
    "year": 2023,
    "paradigm": "joint prediction",
    "backbone": "GPT-style video pretraining",
    "notes": "3.06 of 5 subtasks on CALVIN ABC→D, when prior methods in its table stayed below 1.",
    "url": "https://arxiv.org/abs/2312.13139",
    "slug": "gr-1"
  },
  {
    "name": "GENIMA",
    "org": null,
    "year": 2024,
    "paradigm": "joint prediction",
    "backbone": "Stable Diffusion",
    "notes": "Draws the target actions into the image, so the action is itself a picture.",
    "url": "https://arxiv.org/abs/2407.07875",
    "slug": "genima"
  },
  {
    "name": "Video Prediction Policy",
    "org": null,
    "year": 2024,
    "paradigm": "inverse dynamics",
    "backbone": "Stable Video Diffusion",
    "notes": "Uses predicted video representations to drive the policy.",
    "url": "https://arxiv.org/abs/2412.14803",
    "slug": "video-prediction-policy"
  },
  {
    "name": "Unified Video Action",
    "org": null,
    "year": 2025,
    "paradigm": "joint prediction",
    "backbone": null,
    "notes": "Generates video and actions in one model.",
    "url": "https://arxiv.org/abs/2503.00200",
    "slug": "unified-video-action"
  },
  {
    "name": "mimic-video",
    "org": null,
    "year": 2025,
    "paradigm": "inverse dynamics",
    "backbone": "Cosmos-Predict2",
    "notes": "Video-first policy that reads actions off generated futures.",
    "url": "https://arxiv.org/abs/2512.15692",
    "slug": "mimic-video"
  },
  {
    "name": "Motus",
    "org": null,
    "year": 2025,
    "paradigm": "hybrid",
    "backbone": null,
    "notes": "Mixes the VLA and WAM recipes.",
    "url": "https://arxiv.org/abs/2512.13030",
    "slug": "motus"
  },
  {
    "name": "Cosmos Policy",
    "org": "NVIDIA",
    "year": 2026,
    "paradigm": "joint prediction",
    "backbone": "Cosmos",
    "notes": "Encodes actions as extra latent video frames.",
    "url": "https://arxiv.org/abs/2601.16163",
    "slug": "cosmos-policy"
  },
  {
    "name": "LingBot-VA",
    "org": "Ant Group",
    "year": 2026,
    "paradigm": "inverse dynamics",
    "backbone": "Wan 2.2 (5B)",
    "notes": "Mixture-of-transformers design; 16,000 hours of pretraining.",
    "url": "https://arxiv.org/abs/2601.21998",
    "slug": "lingbot-va"
  },
  {
    "name": "DreamZero",
    "org": "NVIDIA",
    "year": 2026,
    "paradigm": "joint prediction",
    "backbone": "Wan 2.1 (14B)",
    "notes": "1,750 Elo on RoboArena (April 2026 snapshot), ahead of Pi-0.5 at 1,622. Trained only on DROID.",
    "url": "https://arxiv.org/abs/2602.15922",
    "slug": "dreamzero"
  },
  {
    "name": "DreamDojo",
    "org": null,
    "year": 2026,
    "paradigm": "latent actions",
    "backbone": "Cosmos-Predict2.5",
    "notes": "Learns latent actions from video.",
    "url": "https://arxiv.org/abs/2602.06949",
    "slug": "dreamdojo"
  },
  {
    "name": "BagelVLA",
    "org": null,
    "year": 2026,
    "paradigm": "hybrid",
    "backbone": "BAGEL",
    "notes": "A VLA that also predicts visual futures.",
    "url": "https://arxiv.org/abs/2602.09849",
    "slug": "bagelvla"
  },
  {
    "name": "Fast-WAM",
    "org": null,
    "year": 2026,
    "paradigm": "representation only",
    "backbone": "Wan 2.2 (5B)",
    "notes": "Skips video generation at inference. Fig. 4 on one RTX 5090D towel-fold: Fast-WAM 190 ms, Fast-WAM-Joint 580 ms, Fast-WAM-IDM 810 ms, Pi-0.5 180 ms.",
    "url": "https://arxiv.org/abs/2603.16666",
    "slug": "fast-wam"
  },
  {
    "name": "Cortex 2.0",
    "org": "Sereact",
    "year": 2026,
    "paradigm": "hybrid",
    "backbone": null,
    "notes": "Uses a world model for planning.",
    "url": "https://arxiv.org/abs/2604.20246",
    "slug": "cortex-2-0"
  },
  {
    "name": "DVA",
    "org": "Rhoda AI",
    "year": 2026,
    "paradigm": "inverse dynamics",
    "backbone": "Causal video model",
    "notes": "Direct video-action model.",
    "url": "https://www.rhoda.ai/research/direct-video-action",
    "slug": "dva"
  },
  {
    "name": "Being-H0.7",
    "org": "BeingBeyond",
    "year": 2026,
    "paradigm": "latent actions",
    "backbone": null,
    "notes": "Latent world-action model bridging the two recipes.",
    "url": "https://research.beingbeyond.com/being-h07",
    "slug": "being-h0-7"
  },
  {
    "name": "Odyssey-3",
    "org": "Odyssey",
    "year": 2026,
    "paradigm": "representation only",
    "backbone": null,
    "notes": "Foundation world model; Odyssey trains an action decoder on its representations. Research preview as Odyssey-3 Flash; no open weights stated.",
    "url": "https://odyssey.systems/meet-odyssey-3",
    "slug": "odyssey-3"
  }
]