{
  "_comment": "Verbatim `global_task` instruction strings passed to `script/eval_policy.py --global_task` when evaluating the m1_mix checkpoint on each of the five RMBench M1 tasks. Strings are reproduced EXACTLY as used, including original spelling (e.g. 'traies'). For M1 tasks the planner/vLLM is not used; this single global instruction is set directly as the execution-module language input for the whole episode.",
  "model": "Mem-0 execution module (m1_mix) — Qwen3-VL-2B-Instruct backbone + MemoryBank + DiT-B flow-matching action head + subtask-end classifier",
  "checkpoint": "checkpoint/m1_mix_final_step50000.pt",
  "norm_stats": "norm_stats/norm_stats.json",
  "eval_settings": {
    "task_config": "demo_clean",
    "instruction_type": "unseen",
    "num_episodes_per_task": 100,
    "action_horizon": 30,
    "ckpt_setting_label": "m1mix",
    "vllm_required": false
  },
  "tasks": {
    "swap_blocks": {
      "global_task": "There are three traies on the table, and two blocks are placed in two different traies. You may move only one block at a time, and each tray can hold at most one block. Swap the positions of the two blocks. Finally press the button.",
      "success_rate": 0.81,
      "reward": 0.81
    },
    "rearrange_blocks": {
      "global_task": "Move the block between the two mats onto the empty mat, press the button, then move the other block (the one that started on a mat) to the space between the two mats.",
      "success_rate": 0.86,
      "reward": 0.86
    },
    "put_back_block": {
      "global_task": "There are four mats, one block, and a button on the table. One block is on one of the mats. First, put the block to the center, then press the button. Then, put the block back in its original position.",
      "success_rate": 1.0,
      "reward": 1.0
    },
    "swap_T": {
      "global_task": "Swap the poses of the two T-blocks, including both position and orientation.",
      "success_rate": 0.13,
      "reward": 0.13
    },
    "observe_and_pickup": {
      "global_task": "Initially, there is one target object on the shelf and five random objects on the table. Then, a screen obscures the target object. Pick up the corresponding target object from the table and lift it up.",
      "success_rate": 0.03,
      "reward": 0.0
    }
  },
  "average_success_rate": 0.566
}