{
  "niche": "vla-policies",
  "count": 18,
  "records": [
    {
      "name": "RT-1",
      "org": "Google",
      "params": "35M",
      "type": "VLA",
      "training_data": "~130k demos (Everyday Robots)",
      "embodiment": "single-arm",
      "open_weights": true,
      "year": 2022,
      "url": "https://robotics-transformer1.github.io/",
      "confidence": "high",
      "_sources": [
        "https://robotics-transformer1.github.io/"
      ]
    },
    {
      "name": "RT-1-X",
      "org": "Google DeepMind + consortium",
      "params": "35M",
      "type": "VLA",
      "training_data": "Open X-Embodiment (~160k demos)",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2023,
      "url": "https://robotics-transformer-x.github.io/",
      "confidence": "high",
      "_sources": [
        "https://robotics-transformer-x.github.io/"
      ]
    },
    {
      "name": "RT-2",
      "org": "Google DeepMind",
      "params": "up to 55B",
      "type": "VLA",
      "training_data": "web-scale VLM + robot data",
      "embodiment": "single-arm",
      "open_weights": false,
      "year": 2023,
      "url": "https://robotics-transformer2.github.io/",
      "confidence": "high",
      "_sources": [
        "https://robotics-transformer2.github.io/"
      ]
    },
    {
      "name": "RT-2-X",
      "org": "Google DeepMind",
      "params": "55B",
      "type": "VLA",
      "training_data": "Open X-Embodiment",
      "embodiment": "multi",
      "open_weights": false,
      "year": 2023,
      "url": "https://robotics-transformer-x.github.io/",
      "confidence": "medium",
      "_sources": [
        "https://robotics-transformer-x.github.io/"
      ]
    },
    {
      "name": "Octo",
      "org": "Berkeley/Stanford/CMU/GDM",
      "params": "27M (Small) / 93M (Base)",
      "type": "policy",
      "training_data": "800k Open X-Embodiment trajectories",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2024,
      "url": "https://octo-models.github.io/",
      "confidence": "high",
      "_sources": [
        "https://octo-models.github.io/"
      ]
    },
    {
      "name": "OpenVLA",
      "org": "Stanford + consortium",
      "params": "7B",
      "type": "VLA",
      "training_data": "970k Open X-Embodiment episodes",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2024,
      "url": "https://arxiv.org/abs/2406.09246",
      "confidence": "high",
      "_sources": [
        "https://arxiv.org/abs/2406.09246"
      ]
    },
    {
      "name": "pi0 (openpi)",
      "org": "Physical Intelligence",
      "params": "3B (PaliGemma VLM + action expert)",
      "type": "VLA",
      "training_data": "10k+ hrs, 7 platforms, 68 tasks; flow-matching 50 Hz",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2024,
      "url": "https://www.pi.website/blog/openpi",
      "confidence": "high",
      "_sources": [
        "https://www.pi.website/blog/openpi"
      ]
    },
    {
      "name": "pi0.5",
      "org": "Physical Intelligence",
      "params": "~3B",
      "type": "VLA",
      "training_data": "open-world generalization follow-up to pi0",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2025,
      "url": "https://www.pi.website/blog/pi05",
      "confidence": "medium",
      "_sources": [
        "https://www.pi.website/blog/pi05"
      ]
    },
    {
      "name": "GR00T N1",
      "org": "NVIDIA",
      "params": "2B",
      "type": "VLA",
      "training_data": "real trajectories + human video + synthetic",
      "embodiment": "humanoid",
      "open_weights": true,
      "year": 2025,
      "url": "https://arxiv.org/abs/2503.14734",
      "confidence": "high",
      "_sources": [
        "https://arxiv.org/abs/2503.14734"
      ]
    },
    {
      "name": "GR00T N1.5",
      "org": "NVIDIA",
      "params": "3B",
      "type": "VLA",
      "training_data": "+ DreamGen synthetic data",
      "embodiment": "humanoid",
      "open_weights": true,
      "year": 2025,
      "url": "https://research.nvidia.com/labs/gear/gr00t-n1_5/",
      "confidence": "high",
      "_sources": [
        "https://research.nvidia.com/labs/gear/gr00t-n1_5/"
      ]
    },
    {
      "name": "GR00T N1.6",
      "org": "NVIDIA",
      "params": "3B",
      "type": "VLA",
      "training_data": "2x larger DiT + Cosmos-2B VLM + thousands hrs teleop",
      "embodiment": "humanoid (YAM, AgiBot G1, Unitree G1)",
      "open_weights": true,
      "year": 2025,
      "url": "https://research.nvidia.com/labs/gear/gr00t-n1_6/",
      "confidence": "high",
      "_sources": [
        "https://research.nvidia.com/labs/gear/gr00t-n1_6/"
      ]
    },
    {
      "name": "ACT (Action Chunking Transformer)",
      "org": "Stanford (ALOHA)",
      "params": "~80M",
      "type": "policy",
      "training_data": "ALOHA teleoperation demos",
      "embodiment": "bimanual",
      "open_weights": true,
      "year": 2023,
      "url": "https://github.com/tonyzhaozh/act",
      "confidence": "medium",
      "_sources": [
        "https://github.com/tonyzhaozh/act"
      ]
    },
    {
      "name": "Diffusion Policy",
      "org": "Columbia/TRI/MIT (Shuran Song)",
      "params": "small (task-dependent)",
      "type": "policy",
      "training_data": "teleoperation demos",
      "embodiment": "single/bimanual",
      "open_weights": true,
      "year": 2023,
      "url": "https://diffusion-policy.cs.columbia.edu/",
      "confidence": "high",
      "_sources": [
        "https://diffusion-policy.cs.columbia.edu/"
      ]
    },
    {
      "name": "SmolVLA",
      "org": "Hugging Face (LeRobot)",
      "params": "~450M",
      "type": "VLA",
      "training_data": "community LeRobot datasets",
      "embodiment": "multi",
      "open_weights": true,
      "year": 2025,
      "url": "https://huggingface.co/blog/smolvla",
      "confidence": "high",
      "_sources": [
        "https://huggingface.co/blog/smolvla"
      ]
    },
    {
      "name": "DYNA-2",
      "org": "Dyna Robotics",
      "params": "undisclosed",
      "type": "world-action-model",
      "training_data": "1M+ hrs egocentric human video",
      "embodiment": "bimanual (YAM)",
      "open_weights": false,
      "year": 2026,
      "url": "https://www.dyna.co/dyna-2",
      "confidence": "high",
      "_sources": [
        "https://www.dyna.co/dyna-2"
      ]
    },
    {
      "name": "Helix",
      "org": "Figure AI",
      "params": "undisclosed (System1 + System2)",
      "type": "VLA",
      "training_data": "internal multi-operator teleop (~500 hrs)",
      "embodiment": "humanoid (Figure 02)",
      "open_weights": false,
      "year": 2025,
      "url": "https://www.figure.ai/news/helix",
      "confidence": "medium",
      "_sources": [
        "https://www.figure.ai/news/helix"
      ]
    },
    {
      "name": "Redwood",
      "org": "1X Technologies",
      "params": "~160M",
      "type": "VLA",
      "training_data": "teleoperation (Expert Mode)",
      "embodiment": "humanoid (NEO)",
      "open_weights": false,
      "year": 2025,
      "url": "https://www.1x.tech/",
      "confidence": "medium",
      "_sources": [
        "https://www.1x.tech/"
      ]
    },
    {
      "name": "Gemini Robotics",
      "org": "Google DeepMind",
      "params": "undisclosed",
      "type": "VLA",
      "training_data": "Gemini-based; robot fine-tuning",
      "embodiment": "multi",
      "open_weights": false,
      "year": 2025,
      "url": "https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/",
      "confidence": "medium",
      "_sources": [
        "https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/"
      ]
    }
  ],
  "_generated": "data-refinery"
}