JamesK2W's picture
Add 28 tasks (phase 2)
6d0f758 verified
Raw History Blame Contribute Delete
18.8 kB
{
"schema": "efe-task-dag/1",
"taxonomy": "v2",
"taskKey": "behavior-1k/halve_an_egg",
"suite": "behavior-1k",
"base": "behavior-halve-an-egg-i00",
"title_zh": "鸡蛋切半装盘",
"summary_zh": "熟鸡蛋切成两半且两半均由盘面承托、切肉刀位于水槽内部,所有目标对象均离手且线速度不大于0.05米/秒,全部条件共同连续成立30个控制步。",
"embodiment_zh": "两模式均为R1Pro轮式移动双臂机器人,带可动躯干,左右臂各7自由度,各配二指夹爪并使用辅助抓取;头部与双腕相机,底盘可平移与转向。",
"task_class": "compositional",
"analysis_level": "step",
"spec": {
"version": "1.2",
"task": "behavior-halve-an-egg-i00",
"bench": "behavior-1k",
"progress": "live",
"entities": {},
"milestones": [
{
"id": "fridge_opened",
"zh": "打开冰箱",
"mode": "event",
"when": null,
"after": [],
"hold_s": 0,
"goal": false,
"freeze_held": false,
"criterion_zh": "头部或腕部视角可见冰箱白色门板从闭合位置向外转开,露出熟鸡蛋所在的低层搁板,并留出能让鸡蛋通过的开口;仅碰到把手或门板仍贴合柜体不算完成。开门一旦完成即记住,后续关门不撤销。",
"step": {
"cat": "机构",
"item": "打开"
},
"contact": [
{
"who": "手指/夹爪",
"how": "抓握",
"note_zh": "一侧夹爪抓住冰箱门把手,沿门铰链方向向外拉开",
"plan": 1
},
{
"who": "手指/夹爪",
"how": "勾",
"note_zh": "一侧夹爪手指勾住冰箱门把手或门沿向外拉开",
"plan": 2
}
],
"optional": false,
"verify_zh": null,
"sources": [
"inputs/task/privileged/instruction.md:4",
"inputs/bddl/halve_an_egg.bddl:21-29",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1279-1284",
"inputs/oracle/run.json:1,步骤50描述首次开门前冰箱门关闭",
"inputs/oracle/replay.mp4 @ 30秒门闭合、45-60秒门已打开,机器人在厨房内"
],
"conf": 6,
"note_zh": "鸡蛋初始在冰箱内;成功视频在首次拉门前显示闭合门板,之后可见内部搁板。此为默认保留的必要前置步骤,目标不要求最后关门;厨房内无需另开房门。"
},
{
"id": "egg_halved",
"zh": "鸡蛋切半",
"mode": "state",
"when": null,
"after": [
"fridge_opened"
],
"hold_s": 0,
"goal": true,
"freeze_held": false,
"criterion_zh": "头部或腕部视角可见完整熟鸡蛋经切割变为两个独立半蛋,能辨认半蛋的切面或分开的两块;只见刀靠近完整鸡蛋不算,两半任一随后被切成碎块或不再存在即失去此状态。最终联合验收还要求两半均存在且在盘上、刀在水槽内,所有目标对象(两半蛋、盘、刀、水槽)均未被夹爪附着且线速度不大于0.05米/秒,连续30个30赫兹控制步成立;此共同窗口不单独计步。",
"step": {
"cat": "材料",
"item": "切"
},
"contact": [
{
"who": "手持工具",
"how": "压/擦",
"note_zh": "一侧夹爪持切肉刀,用刀刃压切熟鸡蛋;目标接触链只标刀刃与鸡蛋",
"plan": 1
},
{
"who": "手指/夹爪",
"how": "抓握",
"note_zh": "一侧夹爪握住熟鸡蛋,使鸡蛋接触放稳的切肉刀刀刃",
"plan": 2
}
],
"optional": false,
"verify_zh": null,
"sources": [
"inputs/task/privileged/instruction.md:4-9",
"inputs/task/privileged/instruction.md:28",
"inputs/bddl/halve_an_egg.bddl:27-37",
"inputs/oracle/run.json:1,步骤13输出SlicingRule接触与SlicerActive条件",
"inputs/oracle/replay.mp4 @ 189-192秒,刀接触红花盘内鸡蛋并形成两半",
"inputs/grader/efe_behavior-files/rcb_behavior.py:102-104",
"inputs/grader/efe_behavior-files/rcb_behavior.py:150-167",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1414-1434"
],
"conf": 6,
"note_zh": "一次切割同时产生两个real目标,只建一个切割节点;不规定切割地点、用手、姿态或必须先把整蛋放上案板。切半结果可以在夹持中成立,因此关闭持物冻结。两半都留存是实时目标,不能把它写成永不撤销的历史切割事件。"
},
{
"id": "egg_halves_on_plate_1",
"zh": "第1半蛋上盘",
"mode": "state",
"when": null,
"after": [
"egg_halved"
],
"hold_s": 0,
"goal": true,
"freeze_held": true,
"criterion_zh": "头部或腕部视角中,红花圆盘上至少有1个独立半蛋,由盘面承托并与盘接触,半蛋已离开夹爪且静止;悬在盘上方、仍夹在手里或落在盘外不算,盘上满足条件的半蛋少于1个即失去。最终联合验收还要求两半均存在且在盘上、刀在水槽内,所有目标对象(两半蛋、盘、刀、水槽)均未被夹爪附着且线速度不大于0.05米/秒,连续30个30赫兹控制步成立;此共同窗口不单独计步。",
"step": {
"cat": "关系",
"item": "放置"
},
"contact": [
{
"who": "手指/夹爪",
"how": "抓握",
"note_zh": "一侧夹爪抓住一个半蛋并放到盘面上;左右夹爪均可",
"plan": 1
},
{
"who": "底盘",
"how": "支撑",
"note_zh": "轮式底盘支承机器人,在半蛋切割位置与盘子之间换位搬运",
"plan": 1
},
{
"who": "手持工具",
"how": "压/擦",
"note_zh": "一侧夹爪持刀在盘内切开整蛋,两半直接由盘面承接,无须分别抓取",
"plan": 2
}
],
"optional": false,
"verify_zh": null,
"sources": [
"inputs/task/privileged/instruction.md:8",
"inputs/task/privileged/instruction.md:27",
"inputs/bddl/halve_an_egg.bddl:36-37",
"inputs/task/privileged/environment/instance.json:1,盘和案板在相邻台面上",
"inputs/oracle/replay.mp4 @ 175秒整蛋移到盘上、189-192秒盘内切割,两半直接留盘",
"inputs/oracle/run.json:1,步骤205-211记录整蛋先上盘、切割后两半均在盘上",
"inputs/grader/efe_behavior-files/rcb_behavior.py:102-104",
"inputs/grader/efe_behavior-files/rcb_behavior.py:150-167",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1414-1434"
],
"conf": 6,
"note_zh": "按盘上半蛋个数计第1个,不绑定可见身份;两半可同时达成,因此计数依赖不强制不同完成时刻。另一方案是在盘上直接切蛋,刀具接触可同时产生切半与装盘结果,不要求切后再搬运。搬运方案底盘是否必须移动随切割地点而变,按S17保留底盘接触;松手与局部静止是放置验收,全体目标共同30步窗口只作联合成功条件。"
},
{
"id": "egg_halves_on_plate_2",
"zh": "第2半蛋上盘",
"mode": "state",
"when": null,
"after": [
"egg_halves_on_plate_1"
],
"hold_s": 0,
"goal": true,
"freeze_held": true,
"criterion_zh": "头部或腕部视角中,红花圆盘上至少有2个独立半蛋,由盘面承托并与盘接触,半蛋已离开夹爪且静止;悬在盘上方、仍夹在手里或落在盘外不算,盘上满足条件的半蛋少于2个即失去。最终联合验收还要求两半均存在且在盘上、刀在水槽内,所有目标对象(两半蛋、盘、刀、水槽)均未被夹爪附着且线速度不大于0.05米/秒,连续30个30赫兹控制步成立;此共同窗口不单独计步。",
"step": {
"cat": "关系",
"item": "放置"
},
"contact": [
{
"who": "手指/夹爪",
"how": "抓握",
"note_zh": "一侧夹爪抓住一个半蛋并放到盘面上;左右夹爪均可",
"plan": 1
},
{
"who": "底盘",
"how": "支撑",
"note_zh": "轮式底盘支承机器人,在半蛋切割位置与盘子之间换位搬运",
"plan": 1
},
{
"who": "手持工具",
"how": "压/擦",
"note_zh": "一侧夹爪持刀在盘内切开整蛋,两半直接由盘面承接,无须分别抓取",
"plan": 2
}
],
"optional": false,
"verify_zh": null,
"sources": [
"inputs/task/privileged/instruction.md:8",
"inputs/task/privileged/instruction.md:27",
"inputs/bddl/halve_an_egg.bddl:36-37",
"inputs/task/privileged/environment/instance.json:1,盘和案板在相邻台面上",
"inputs/oracle/replay.mp4 @ 175秒整蛋移到盘上、189-192秒盘内切割,两半直接留盘",
"inputs/oracle/run.json:1,步骤205-211记录整蛋先上盘、切割后两半均在盘上",
"inputs/grader/efe_behavior-files/rcb_behavior.py:102-104",
"inputs/grader/efe_behavior-files/rcb_behavior.py:150-167",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1414-1434"
],
"conf": 6,
"note_zh": "按盘上半蛋个数计第2个,不绑定可见身份;两半可同时达成,因此计数依赖不强制不同完成时刻。另一方案是在盘上直接切蛋,刀具接触可同时产生切半与装盘结果,不要求切后再搬运。搬运方案底盘是否必须移动随切割地点而变,按S17保留底盘接触;松手与局部静止是放置验收,全体目标共同30步窗口只作联合成功条件。"
},
{
"id": "knife_in_sink",
"zh": "刀具入水槽",
"mode": "state",
"when": null,
"after": [],
"hold_s": 0,
"goal": true,
"freeze_held": true,
"criterion_zh": "头部或腕部视角可见切肉刀放在水槽凹入的盆腔中,刀的包围盒中心位于水槽包围盒及内部容积内,夹爪已放开且刀静止;搁在水槽旁的沥水台、横在盆沿上而中心未入内、仍拿着刀均不算,刀离开盆腔即失去。最终联合验收还要求两半均存在且在盘上、刀在水槽内,所有目标对象(两半蛋、盘、刀、水槽)均未被夹爪附着且线速度不大于0.05米/秒,连续30个30赫兹控制步成立;此共同窗口不单独计步。",
"step": {
"cat": "关系",
"item": "放入"
},
"contact": [
{
"who": "手指/夹爪",
"how": "抓握",
"note_zh": "一侧夹爪握住切肉刀刀柄,搬到水槽盆腔内后放开",
"plan": 1
},
{
"who": "底盘",
"how": "支撑",
"note_zh": "轮式底盘支承机器人持刀从案板或盘子一侧移到水槽盆腔前",
"plan": 1
}
],
"optional": false,
"verify_zh": null,
"sources": [
"inputs/task/privileged/instruction.md:9",
"inputs/task/privileged/instruction.md:26",
"inputs/bddl/halve_an_egg.bddl:38",
"inputs/oracle/replay.mp4 @ 195-207秒,刀先落沥水台再移入水槽盆腔",
"inputs/oracle/run.json:1,步骤215-219记录刀先落沥水台、随后移入盆腔",
"inputs/grader/efe_behavior-files/rcb_behavior.py:102-104",
"inputs/grader/efe_behavior-files/rcb_behavior.py:150-167",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1414-1434"
],
"conf": 6,
"note_zh": "无须把刀入水槽强制排在装盘之后;评分器只检查最终inside关系,没有原生阶段顺序。若先放刀再取用,state会真实失去并可恢复。"
}
],
"constraints": [
{
"id": "robot_fallen",
"zh": "机器人跌倒",
"never": null,
"criterion_zh": "机器人基座自身向上轴的世界竖直分量小于0.75(倾斜超过约41.4度),或基座比场景地面低超过0.03米;上述或条件连续成立300个控制步(10秒)触发跌倒失败,短暂倾斜后恢复不算。",
"consequence": "立即终止",
"sources": [
"inputs/grader/efe_behavior-files/rcb_behavior.py:92-144",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1483-1488"
]
},
{
"id": "simulator_breakdown",
"zh": "仿真状态失控",
"never": null,
"criterion_zh": "机器人位置、四元数、线速度或关节位置出现非有限数,任一位置坐标绝对值超过10000米,或基座线速度超过100米/秒,立即判仿真失控;每30个控制步还检查目标对象位置,非有限或任一坐标绝对值超过10000米也触发失败。",
"consequence": "立即终止",
"sources": [
"inputs/grader/efe_behavior-files/rcb_behavior.py:106-113",
"inputs/grader/efe_behavior-files/rcb_behavior.py:1437-1471"
]
}
],
"success": null,
"atoms": [
{
"atom": "fridge_open",
"level": "new_code",
"source": "proc.electric_refrigerator.n.01_1.open;需补充冰箱门开合状态逐帧记录,数值门槛应沿用模拟器Open对象状态"
},
{
"atom": "egg_halves_real",
"level": "new_code",
"source": "flags.lit:(real half__hard-boiled_egg.n.01_1) 与 flags.lit:(real half__hard-boiled_egg.n.01_2);需补录逐帧目标字面量"
},
{
"atom": "egg_halves_on_plate_count",
"level": "new_code",
"source": "flags.lit:(ontop half__hard-boiled_egg.n.01_1 plate.n.04_1) 与 flags.lit:(ontop half__hard-boiled_egg.n.01_2 plate.n.04_1);需补录后按成立数量计数"
},
{
"atom": "knife_inside_sink",
"level": "new_code",
"source": "flags.lit:(inside carving_knife.n.01_1 sink.n.01_1);需补录逐帧目标字面量"
},
{
"atom": "goal_objects_released_and_still",
"level": "new_code",
"source": "_success_step 的 goal_objects、_ag_obj_in_hand 与 get_linear_velocity;需补录两半蛋、盘、刀、水槽的附着及线速度,用于放置验收及全体联合成功"
},
{
"atom": "success_rule",
"level": "new_code",
"source": "Session._rest.done 与 _rest.run;需补录全体目标、松手、线速度不大于0.05米/秒共同连续30步的成功状态"
},
{
"atom": "robot_fallen",
"level": "new_code",
"source": "Session._fall.fallen;需补录满足300控制步窗口的跌倒状态"
},
{
"atom": "simulator_breakdown",
"level": "new_code",
"source": "Session.breakdown;需补录失控状态"
}
],
"notes_zh": "联合成功条件为BDDL五个目标字面量全部成立,并对两半蛋、盘、刀、水槽共同要求未被夹爪附着且线速度不大于0.05米/秒,连续30控制步;没有日志,success保持null,不能仅凭几何目标替代完整成功规则。各节点hold_s为0,全体共同窗口不重复加到各个目标上。"
},
"native_metrics": [
{
"name": "q_score",
"unit": "1",
"threshold": 1,
"direction": "higher_better"
}
],
"open_questions_zh": [
"输入states/index.json的logs为空,oracle也无状态日志;按DECISIONS.md P8,每个when、success及约束never均为null,atoms列出需要新增记录的状态,不能验证逐帧完成时刻或success与评分器的一致性。",
"按DECISIONS.md D2/Y17核查工具拿取:文本及成功运行步骤13所示SlicingRule只要求可切对象接触活跃刀刃,不要求刀被夹爪持有;没有证据证明所有合法路线必须先拿刀,故不设拿刀节点,并保留移动鸡蛋接触放稳刀刃的接触方案。",
"按DECISIONS.md D5,文本只说明整蛋起初在冰箱,没有独立要求或评分鸡蛋取出、放在案板;取蛋并入切蛋或装盘的过程,不单列取出节点。",
"按DECISIONS.md D10及CASES.md OPEN-21,切蛋采用当前暂定解读:一次切割同时产生两个real目标,只建一个切半节点;按N4,两半上同一盘的目标以第1、第2半蛋计数,可在盘上切割时同时完成。",
"模式的世界目标、松手静置规则及跌倒规则相同;privileged评分器还要求两次回放均成功、最终状态哈希相同且初始实例匹配(privileged/grade.py:168-174),standard直接使用服务的live_success及finalized等有效性字段(standard/grade.py:29-36)。这些交付与回放有效性检查不作为世界步骤或语义节点条件,success因无状态日志保持null。",
"冻结instance.json没有记录冰箱门关节,BDDL也没有显式open初值;首次开门前的oracle视频30秒及运行步骤50支持门关闭,故保留开门前置但不虚构开度角阈值。"
],
"conf": 6,
"conf_reason_zh": "四个目标步骤直接来自两模式文本与五个BDDL字面量,成功视频支持关闭冰箱的前置开门及在盘内切蛋方案;没有任何状态日志,无法执行条件绑定、精确静置检查或跨轨迹一致性验证。",
"review": {
"status": "agreed",
"rounds": 1,
"reviewers": [
{
"engine": "claude",
"model": "claude-sonnet-5-5",
"round": 1,
"verdict": "agree",
"conf": 6,
"summary_zh": "无状态日志,只能对照文本、BDDL与评分器核对:步骤集合(开冰箱前置、切半、第1/2半上盘计数、刀入水槽)与目标一致,边无习惯边,约束为跌倒与仿真失控(立即终止),when/success按P8为null并已在open_questions_zh说明;拿刀不设节点的理由成立(刀本就要放入水槽,切割只需蛋接触刀刃)。未发现需改内容的错误。",
"n_issues": 0,
"n_major": 0
}
],
"open_zh": []
},
"evidence_check": {
"n_state_runs": 0,
"n_faithful": 0,
"success_runs_reaching_all": "0/0",
"spec_success_eq_grader": "0/0",
"diverged_runs": [],
"notes_zh": [
"when=null 的节点不在状态日志上检查:['fridge_opened', 'egg_halved', 'egg_halves_on_plate_1', 'egg_halves_on_plate_2', 'knife_in_sink']"
]
},
"provenance": {
"prompt": "prompts/01_task_dag.md",
"prompt_sha256": "b8e79c84b73950529a4fa6b988412b2d221d5df4e052adafcd9605afac64aa25",
"engine": "codex",
"model": "gpt-6.1-sol",
"effort": "high",
"created": "2026-10-29T09:43:10Z",
"review_prompt": "prompts/02_task_dag_review.md",
"review_prompt_sha256": "305c2c5feb2340e0c426ebfe883251f74824cbde2c3e230baa9c86fc05c31550",
"reconcile_prompt": "prompts/05_reconcile.md",
"reconcile_prompt_sha256": "8c827af3ae3f77d09229399fb41283c50a8f725302cbe3b263b8051213dde2b2",
"calls": [
{
"stage": "dag",
"round": 0,
"ok": true,
"space": "work/tasks/behavior-1k__halve_an_egg/01_dag"
},
{
"stage": "dag_review",
"round": 1,
"ok": true,
"space": "work/tasks/behavior-1k__halve_an_egg/02_review_r1"
}
],
"review_engine": "claude",
"review_model": "claude-sonnet-5-5",
"inputs_manifest_sha256": "442b2e818233aa46f6dc86eebca507eab22b6631fd9b0be980a0732cae5c08cf",
"repo_commit": "80cac7a5a227db930c659c3307a1e79c2a84bc53"
}
}