{
  "main": [
    {
      "title": "Visual reasoning",
      "metric": "8-benchmark mean accuracy (%)",
      "labels": [
        "Qwen2.5-VL\nmatched base",
        "Uni-LaDiR"
      ],
      "values": [
        63.55,
        75.55
      ],
      "gain": "+18.9% relative"
    },
    {
      "title": "Visual math & logic",
      "metric": "4-benchmark mean accuracy (%)",
      "labels": [
        "ILVR\nreported baseline",
        "Uni-LaDiR"
      ],
      "values": [
        45.68,
        49.03
      ],
      "gain": "+7.3% relative"
    },
    {
      "title": "Robot manipulation",
      "metric": "RLBench mean success (%)",
      "labels": [
        "LaST\u2080",
        "Uni-LaDiR"
      ],
      "values": [
        82,
        87
      ],
      "gain": "+6.1% relative"
    }
  ],
  "sharing": [
    {
      "title": "Text + image teachers",
      "metric": "4-benchmark mean accuracy (%)",
      "labels": [
        "Separate\nspaces",
        "Text + image\nshared"
      ],
      "values": [
        62.97,
        67.66
      ],
      "gain": "+7.4% relative"
    },
    {
      "title": "Visual + 3D + state teachers",
      "metric": "LIBERO / RLBench mean success (%)",
      "labels": [
        "Separate",
        "Visual\n+ 3D",
        "Visual\n+ state",
        "3D\n+ state",
        "All three"
      ],
      "values": [
        90.05,
        90.6,
        90.25,
        91.25,
        93.08
      ],
      "gain": "+3.4% relative"
    }
  ]
}