{
 "generated": "2026-11-20",
 "sha": "",
 "node": {
  "date": "2026-11-05",
  "updated": "2026-11-06",
  "kind": "topic",
  "category": "评测与实验",
  "group": "probes",
  "page": "/deepdive/robodojo-robot-goals-llm-entries/",
  "summary": "RoboDojo 榜单 7 月到 11 月：最好成功率从 8.80% 涨到 Agent 轨的 68.87%（人类专家 76.03%）；LLM 直接当策略、审 VLA、写代码冻结、整套 Agent 轨四个入口，精度 4.00% 与真机 1/33 是它们都还没过的坎",
  "version": 2,
  "tags": [
   "RoboDojo",
   "机器人操作",
   "VLA",
   "世界动作模型",
   "LLM 当策略",
   "Code-as-Policies",
   "GPT-6 Astra",
   "Agent 轨",
   "XPolicyLab",
   "具身智能评测"
  ],
  "langs": [
   "zh",
   "en"
  ],
  "words": 11087,
  "n_sections": 9,
  "dd_id": "DD·0168",
  "related": [
   "autonomous-experiments-gap-map",
   "andon-labs-pion",
   "llm-pollution-media-chain",
   "ai-jargon-index"
  ],
  "id": "dd:robodojo-robot-goals-llm-entries",
  "type": "Article",
  "article": "robodojo-robot-goals-llm-entries",
  "name": "机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
  "title": "机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
  "slug": "robodojo-robot-goals-llm-entries",
  "kind_zh": "热点专题",
  "entry": "agent",
  "n_ents": 2,
  "n_entities": 2
 },
 "page": "https://xiaopingfeng.com/graph/e/dd/robodojo-robot-goals-llm-entries/",
 "out": [
  {
   "type": "covers",
   "sentence": "机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬讲到GPT-6 Astra",
   "direction": "out",
   "id": "model:gpt-6-astra",
   "name": "GPT-6 Astra",
   "role": "mention",
   "status": "agent",
   "evidence": []
  },
  {
   "type": "covers",
   "sentence": "机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬讲到Einsia",
   "direction": "out",
   "id": "org:einsia-ai",
   "name": "Einsia",
   "role": "mention",
   "status": "agent",
   "evidence": []
  }
 ],
 "in": [
  {
   "type": "part_of",
   "sentence": "从 22% 到 69%，都在仿真里属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-tldr",
   "name": "从 22% 到 69%，都在仿真里",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "这座山 是怎么量的属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-01",
   "name": "这座山 是怎么量的",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "LLM 从哪几个 口子进来属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-02",
   "name": "LLM 从哪几个 口子进来",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "直接让 LLM 当策略属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-03",
   "name": "直接让 LLM 当策略",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "让 LLM 审别人的动作属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-04",
   "name": "让 LLM 审别人的动作",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "让 LLM 写代码，写完冻结属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-05",
   "name": "让 LLM 写代码，写完冻结",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "整个 Agent 轨属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-06",
   "name": "整个 Agent 轨",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "三个可以被 打脸的判断属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-07",
   "name": "三个可以被 打脸的判断",
   "status": "derived",
   "evidence": []
  },
  {
   "type": "part_of",
   "sentence": "边界和没核到的属于机器人在爬一座“不偏科”的山，LLM 从四个口子往上爬",
   "direction": "in",
   "id": "ddsec:robodojo-robot-goals-llm-entries-sec-08",
   "name": "边界和没核到的",
   "status": "derived",
   "evidence": []
  }
 ]
}