robo-si已验收任务accepted tasks
robo-si / FT

FT · 易碎烧瓶搬运：按身体选工具与能力自建模Fragile flask transport: body-specific tool choice and self-modeling

把三只易碎烧瓶从桌 A 完好送到桌 B 标记区，松手后由桌 B 直接支撑并稳定；不给流程、不指定工具。同一目标与工具库（T1 运输盒、T2 防震套、T3 推车、T4 车载托架、T5 领圈）下，G1＋Dex3（固定或行走）、G1＋Inspire 五指手、Galaxea R1 与 Stretch 3 需按自身能力选择直接搬运或工具。L1 能力全公开，L2 工具相容性不公开，L3 隐藏部分物理量并提供试验瓶，另有两个全路线不可行的退出对照。Deliver three fragile flasks intact from table A to the marked area on table B, released and stable with table B supporting them; no procedure or tool is prescribed. With the same goal and tool library (T1 carrier, T2 sleeve, T3 cart, T4 base racks, T5 collar), G1 + Dex3 (fixed or walking), G1 + Inspire five-finger hands, Galaxea R1 and Stretch 3 must choose direct carrying or tools according to their own abilities. L1 discloses everything, L2 hides tool compatibility, L3 hides physical quantities and supplies a test flask, and two controls are infeasible on every route and require exiting.

验收范围Accepted scope

18/18 项按当前授权修订目标签收，包括 0.1.0 原版（2026-10-27 批准发布）和两项正确退出对照。其中 B.G、L3-3.S 为整批预装运输，C.S 为分段计时，L3-1.G 为调整布局并改用共轴 T5，L3-4.G 只是推车探测与安全返回子任务，不是原始三瓶运输；这些原目标（逐瓶装载与时间预算区分、L3-3“名义破损／慢速完好”前提、L3-4 运到 B）仍未确立。全部为读取真值的专家参考，正式 benchmark 准入仍为 false。All 18/18 cases are signed off against their currently authorized revised targets, including the 0.1.0 original (release approved 2026-10-27) and the two correct-exit controls. B.G and L3-3.S are prepacked shipments, C.S uses a phase budget, L3-1.G uses a revised layout with a coaxial T5, and L3-4.G is only a cart-probe and safe-return component, not the original three-flask transport; those original goals (per-flask loading with time-budget separation, the L3-3 nominal-break/slow-success premise, L3-4 delivery to B) remain unestablished. All are privileged expert references, and formal benchmark admission remains false.

变化的条件轴Condition axes

身体（G 固定／行走、F、W、S）；难度层 L1–L3（能力卡与工具相容性公开程度；隐藏物理量：湿滑瓶面、老化缓冲座、推车能否越槛）；路线场景（短程平地、约 4–8 m 带 90° 转弯和 4 mm 地缝的长程、台阶平台不可达）；在场工具组合与公开时间预算。Body (G fixed/walking, F, W, S); difficulty layer L1–L3 (how much of the capability card and tool compatibility is public; hidden quantities: wet flask surface, aged cushion seats, cart passability over a threshold); route scenario (short flat route, a roughly 4–8 m route with a 90° turn and a 4 mm seam, unreachable step platform); which tools are present and the public time budget.

Embodiments

  • G1＋Dex3（固定骨盆：ORIG、A.G、L3-1.G；可行走：B.G、L3-4.G）G1 + Dex3 (fixed pelvis for ORIG, A.G and L3-1.G; walking for B.G and L3-4.G)
  • G1＋Inspire DFQ 五指手（固定骨盆）G1 + Inspire DFQ five-finger hands (fixed pelvis)
  • Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
  • Stretch 3（移动底盘＋单伸缩臂）Stretch 3 (mobile base with one telescoping arm)

统计Counts

18 个已验收任务／配置，其中 2 个拒答；18 个有倍速视频。18 accepted tasks/configurations, 2 of them refusal cases; 18 with sped-up videos.

证据备份Evidence backup

本族 18/18 个已验收配置的参考证据已归档：7,110 个文件，10.0 GiB。Reference evidence for 18/18 accepted configurations is archived: 7,110 files, 10.0 GiB. 2026-11-03 对远端逐文件核对大小与 SHA-256。Every remote file's size and SHA-256 were checked on 2026-11-03.

FT-ORIG.GFT-L1.FFT-L1.WFT-L1.SFT-A.GFT-A.SFT-B.GFT-B.WFT-B.SFT-C.WFT-C.SFT-L3-1.GFT-L3-1.FFT-L3-2.WFT-L3-3.SFT-L3-4.GFT-I01.WFT-I01.S
8×
8 倍速 · 原始 552 s，播放 69 s8x speed · original 552 s, plays in 69 s

FT-ORIG.G原版：固定骨盆 G1 运输三瓶Original: fixed-pelvis G1 three-flask transport

Unitree G1执行任务Solve the task
条件Conditions
L1 基础层，冻结的 0.1.0 原版；固定骨盆 G1＋Dex3；两桌各 0.62×0.50×0.74 m、间隙 0.12 m，三只 0.12 kg 原版烧瓶（无颈环）；只提供 T1；能力公开；沿用 0.1.0 损伤规则；预算 900 s。L1 baseline, the frozen 0.1.0 original; fixed-pelvis G1 + Dex3; two 0.62×0.50×0.74 m tables with a 0.12 m gap and three 0.12 kg original flasks (no neck ring); only T1 is provided; capabilities public; 0.1.0 damage rule; 900 s budget.
任务设计Task design
把三只烧瓶从桌 A 完好送到桌 B 标记区：离开盒子、由桌 B 直接支撑、松手后稳定 2 秒；装箱可选，本身不计分。参考专家先把运输盒搬近、装入三瓶、双手抓把手搬运，放稳后逐个卸货并松手。Deliver all three flasks intact from table A to the marked area on table B: out of the carrier, directly supported by table B and stable for 2 s after release; packing is optional and earns nothing by itself. The reference expert pulls the carrier closer, loads the flasks, carries it by both handles, sets it down and unloads the flasks one by one.
工具Tools
T1 弹簧阻尼运输盒：三座、双侧开放把手、0.425 kgT1 spring-damper carrier: three seats, two open handles, 0.425 kg
Embodiment
G1＋Dex3，固定骨盆G1 + Dex3, fixed pelvis
是否拒答Refusal
否No
见证结果Witness result
成功：0.1.0 发布验证中三瓶全新运行 3/3（552.11–567.04 s）、单瓶 1/1、独立动作回放 2/2 全部通过，2026-10-27 获作者批准正式发布。见证录像为 multi-1（552.11 s）的独立动作回放；这是读取真值的设计者参考，随机场景成功率与隔离沙盒未验证。Success: the 0.1.0 release validation passed three fresh three-flask runs 3/3 (552.11–567.04 s), a single-flask run 1/1 and independent action replays 2/2, and the author approved the formal release on 2026-10-27. The witness video is the independent action replay of multi-1 (552.11 s); this is a privileged designer reference, and randomized success rates and an isolated sandbox are not verified.
8×
8 倍速 · 原始 169 s，播放 21 s8x speed · original 169 s, plays in 21 s

FT-L1.F换五指手：Inspire 直接逐瓶搬运Five-finger hand: Inspire direct per-flask transfer

Unitree G1执行任务Solve the task
条件Conditions
L1；固定骨盆 G1＋Inspire DFQ 五指手；与 ORIG 同一世界与道具，烧瓶换为带 27 mm 颈环的扩展瓶；能力与工具规格全部公开；FT v0.2 压碎／冲击损伤规则；时限 900 s。L1; fixed-pelvis G1 + Inspire DFQ five-finger hands; same world and props as ORIG, with the extension flask that has a 27 mm neck ring; capabilities and tool specs fully public; FT v0.2 crush/impact damage rule; 900 s limit.
任务设计Task design
只换末端，考察五指手的抓取方式与握力控制。参考解不用工具：右手五指包握瓶颈、拇指与食指勾在颈环下方，抬起后转腰越过运输盒，逐瓶放入桌 B 目标区并松手。Only the end effector changes, testing the five-finger grasp choice and grip-force control. The reference uses no tool: the right hand wraps the neck with five fingers, thumb and index hooked under the ring, lifts, sweeps the waist over the carrier and places each flask in the table-B target before releasing.
工具Tools
T1 弹簧阻尼运输盒（在场，参考解未使用）T1 spring-damper carrier (present, unused by the reference)
Embodiment
G1＋Inspire DFQ 五指手，固定骨盆G1 + Inspire DFQ five-finger hands, fixed pelvis
是否拒答Refusal
否No
见证结果Witness result
成功：一次 169.16 s 的连续运行中，153.8 s 时三瓶完好、直立、在目标区内由桌 B 支撑并稳定。仅一次见证；Inspire 能否用 T1 把手搬运未由此证明。Success: in one continuous 169.16 s run, all three flasks were intact, upright, inside the target, supported by table B and stable at 153.8 s. Single witness only; it does not establish that the Inspire hand can carry T1 by its handles.
8×
8 倍速 · 原始 402 s，播放 50 s8x speed · original 402 s, plays in 50 s

FT-L1.WR1 轮式双臂短距转运R1 wheeled dual-arm short transfer

Galaxea R1执行任务Solve the task
条件Conditions
L1；Galaxea R1 轮式双臂；两桌相距 1.5 m、平地；只有 T1 可见；能力与工具公开；FT v0.2 损伤规则；时限 900 s。L1; Galaxea R1 wheeled dual-arm robot; tables 1.5 m apart on flat floor; only T1 visible; capabilities and tools public; FT v0.2 damage rule; 900 s limit.
任务设计Task design
考察底盘移动中保持载荷稳定，双臂持 T1 或逐瓶搬运均合法。参考解不用 T1：两臂分别在颈环下方侧向夹瓶，分两趟直接运送三瓶。Tests keeping the load steady while the base moves; carrying T1 with both arms or moving flasks individually are both legal. The reference skips T1: each arm side-grasps a flask under its neck ring, delivering the three flasks directly in two trips.
工具Tools
T1 弹簧阻尼运输盒（在场，参考解未使用）T1 spring-damper carrier (present, unused by the reference)
Embodiment
Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
是否拒答Refusal
否No
见证结果Witness result
成功：387.87 s 时三瓶全部交付并稳定，一直保持到记录结束（401.6 s）；无跌倒。Success: all three flasks were delivered and stable at 387.87 s and stayed so until the recording ended (401.6 s); no fall.
8×
8 倍速 · 原始 168 s，播放 21 s8x speed · original 168 s, plays in 21 s

FT-L1.SStretch 单臂逐瓶转运Stretch single-arm per-flask transfer

Stretch 3执行任务Solve the task
条件Conditions
L1；Stretch 3 单伸缩臂；两桌之间 1.5 m 通道；T1、三个 T2 与 T4-S 可见，能力卡写明单臂；FT v0.2 损伤规则；时限 420 s。L1; Stretch 3 with one telescoping arm; a 1.5 m lane between the tables; T1, three T2 and T4-S visible, and the capability card states a single arm; FT v0.2 damage rule; 420 s limit.
任务设计Task design
考察单臂下的任务分解：逐瓶搬运、装 T2 分批或先装 T4-S 统一转运均可。参考解逐瓶夹住瓶颈、沿通道运到桌 B 目标槽，工具均未使用。Tests task decomposition with one arm: per-flask carries, batches in T2 or loading T4-S for one trip are all allowed. The reference pinches each flask by the neck and drives it along the lane to its table-B slot, using no tool.
工具Tools
T1 弹簧阻尼运输盒（在场，参考解未使用）T1 spring-damper carrier (present, unused by the reference)T2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
否No
见证结果Witness result
成功：三瓶全部放入目标区，自 161.6 s 起稳定满 2 s，记录共 168.22 s。Success: all three flasks placed in the target and stable for 2 s from 161.6 s; recording length 168.22 s.
8×
8 倍速 · 原始 449 s，播放 56 s8x speed · original 449 s, plays in 56 s

FT-A.G平坦短程·固定骨盆 G1（工具发现）Flat short route, fixed-pelvis G1 (tool discovery)

Unitree G1执行任务Solve the task
条件Conditions
L2-A；固定骨盆 G1＋Dex3；沿用 ORIG 布局，扩展瓶；T1、T2、T3、T4-W、T4-S、T5 全部在场，几何与接口公开、相容性不公开；FT v0.2 损伤规则；时限 900 s。L2-A; fixed-pelvis G1 + Dex3; ORIG layout with extension flasks; T1, T2, T3, T4-W, T4-S and T5 all present, with geometry and interfaces public but compatibility undisclosed; FT v0.2 damage rule; 900 s limit.
任务设计Task design
判断直接搬运是否优于装箱，即不滥用工具。固定骨盆单臂够不到从桌 A 到目标区的整段路程，参考解不用工具而在桌 B 中转：右手送到中转点、左手再放入目标区，共六次颈部抓取。Tests whether direct carrying beats packing, i.e. not overusing tools. One arm of the fixed-pelvis G1 cannot reach the whole way from table A to the target, so the tool-free reference relays on table B: the right hand stages each flask and the left hand places it in the target, six neck grasps in total.
工具Tools
T1 弹簧阻尼运输盒（未使用）T1 spring-damper carrier (unused)T2 单瓶防震套T2 single-flask cushioned sleeveT3 推车T3 cartT4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)T5 马蹄形领圈：从颈环下方托起烧瓶T5 horseshoe collar: lifts a flask under its neck ring
Embodiment
G1＋Dex3，固定骨盆G1 + Dex3, fixed pelvis
是否拒答Refusal
否No
见证结果Witness result
成功：448.45 s 时三瓶经右手中转、左手交付，全部完好并稳定在目标区；左手每次回到预抓姿态都靠关节运动完成，没有状态重置。Success: at 448.45 s all three flasks, relayed right hand to left hand, were intact and stable in the target; every return of the left hand to its pre-grasp pose was done by joint motion, with no state reset.
8×
8 倍速 · 原始 156 s，播放 19 s8x speed · original 156 s, plays in 19 s

FT-A.S平坦短程·Stretch（工具发现）Flat short route, Stretch (tool discovery)

Stretch 3执行任务Solve the task
条件Conditions
L2-A；Stretch 3；两桌间距按单次转身可达标定；T1、三个 T2、T4-S、T4-W、T3、T5 全部在场，相容性不公开；FT v0.2 损伤规则；时限 360 s。L2-A; Stretch 3; table spacing calibrated so one turn reaches both; T1, three T2, T4-S, T4-W, T3 and T5 all present with compatibility undisclosed; FT v0.2 damage rule; 360 s limit.
任务设计Task design
同样考察是否需要工具。参考解每次夹住一只瓶颈，原地转 180° 直接放入桌 B 目标槽，再转回取下一瓶。Again tests whether a tool is needed. The reference pinches one flask neck at a time, turns 180° in place, places it directly in its table-B slot and turns back for the next.
工具Tools
T1 弹簧阻尼运输盒（未使用）T1 spring-damper carrier (unused)T2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)T4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T3 推车T3 cartT5 马蹄形领圈：从颈环下方托起烧瓶T5 horseshoe collar: lifts a flask under its neck ring
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
否No
见证结果Witness result
成功：三瓶全部交付，自 140.4 s 起稳定满 2 s，记录共 155.46 s。Success: all three flasks delivered and stable for 2 s from 140.4 s; recording length 155.46 s.
8×
8 倍速 · 原始 144 s，播放 18 s8x speed · original 144 s, plays in 18 s

FT-B.G长程转运·行走 G1（整批预装修订版）Long route, walking G1 (prepacked-shipment revision)

Unitree G1执行任务Solve the task
条件Conditions
L2-B；可行走 G1＋Dex3（GR00T 下肢控制）；带 90° 转弯和 4 mm 地缝的 L 形路线；作者授权修订：三瓶初始已装在 T1 座内，全程整托运输，总上限 600 s，取代原逐瓶装卸与时间预算区分要求；其余工具在场。L2-B; walking G1 + Dex3 (GR00T leg control); an L-shaped route with a 90° turn and a 4 mm floor seam; author-approved revision: the three flasks start seated in T1 and travel as one packed shipment, with a 600 s cap, replacing the original per-flask loading and time-budget separation; the other tools are present.
任务设计Task design
双手抓 T1 把手走完全程，把整个运输盒放进桌 B 标记区并松手：盒体由桌 B 直接支撑，三瓶完好、直立地留在座内并稳定 2 秒。修订版 T1 改用短金属立柱把手，原瓶、座位弹簧阻尼与损伤阈值不变。Carry T1 by both handles along the whole route, set the entire carrier inside the table-B marked area and release it: the carrier rests directly on table B and the three flasks stay intact and upright in their seats, stable for 2 s. The revised T1 uses short metal post handles; flasks, seat springs/dampers and damage thresholds are unchanged.
工具Tools
T1 运输盒（预装三瓶，短金属立柱把手）T1 carrier (prepacked with three flasks, short metal post handles)T2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T3 推车T3 cartT4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)
Embodiment
G1＋Dex3，可行走G1 + Dex3, walking
是否拒答Refusal
否No
见证结果Witness result
成功（修订目标）：143.9 s 走完全程并越过地缝，整个运输盒在原 B 标记区内松手、直接支撑并稳定 2 s，三瓶完整直立。原“从桌 A 装瓶并以时间预算区分工具路线”的目标仍未确立。Success (revised goal): at 143.9 s, after the full route and the seam, the whole carrier was released inside the original table-B marker, directly supported and stable for 2 s, with all three flasks intact and upright. The original goal of loading from table A with a time budget that separates tool routes remains unestablished.
8×
8 倍速 · 原始 452 s，播放 57 s8x speed · original 452 s, plays in 57 s

FT-B.W长程转运·R1（T1 双臂搬运）Long route, R1 (two-arm T1 carry)

Galaxea R1执行任务Solve the task
条件Conditions
L2-B；Galaxea R1；带 90° 转弯和 4 mm 地缝的长路线；T1、T2、T4-W、T3、T5 在场，相容性不公开；运行前冻结的公开时间预算 560 s，使逐瓶往返放不进预算；FT v0.2 损伤规则。L2-B; Galaxea R1; a long route with a 90° turn and a 4 mm floor seam; T1, T2, T4-W, T3 and T5 present with compatibility undisclosed; a public 560 s time budget frozen before the run so that per-flask round trips do not fit; FT v0.2 damage rule.
任务设计Task design
把路线风险与时间约束对应到自己的运输方式，而不是固定用某个工具。参考解把三瓶装入 T1，双臂抬盒越过地缝和转弯，放到桌 B 旁后逐瓶卸到目标区。Map route risk and the time limit onto a transport method instead of defaulting to one tool. The reference loads the three flasks into T1, carries it with both arms over the seam and around the turn, sets it beside table B and unloads each flask into the target.
工具Tools
T1 弹簧阻尼运输盒：三座、双侧开放把手T1 spring-damper carrier: three seats, two open handlesT2 单瓶防震套T2 single-flask cushioned sleeveT4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T3 推车T3 cartT5 马蹄形领圈：从颈环下方托起烧瓶T5 horseshoe collar: lifts a flask under its neck ring
Embodiment
Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
是否拒答Refusal
否No
见证结果Witness result
成功：437.935 s 时三瓶完好、直立、松手并稳定在桌 B，在 560 s 预算内；实测逐瓶手持基线需 662.70 s。时间比较只针对所声明的控制器。Success: at 437.935 s all three flasks were intact, upright, released and stable on table B, within the 560 s budget; the measured per-flask hand baseline takes 662.70 s. The timing comparison concerns the declared controllers only.
8×
8 倍速 · 原始 215 s，播放 27 s8x speed · original 215 s, plays in 27 s

FT-B.S长程转运·Stretch（托架加手持一趟运完）Long route, Stretch (rack plus hand carry in one trip)

Stretch 3执行任务Solve the task
条件Conditions
L2-B；Stretch 3；8 m 路线含 90° 转弯和 4 mm 地缝；T1、三个 T2、T4-S 在场；运行前冻结的公开时间预算 240 s，逐瓶往返放不进预算；FT v0.2 损伤规则。L2-B; Stretch 3; an 8 m route with a 90° turn and a 4 mm floor seam; T1, three T2 and T4-S present; a public 240 s budget frozen before the run, which per-flask round trips cannot meet; FT v0.2 damage rule.
任务设计Task design
单臂身体须选择一趟运完的方式。参考解把两只瓶放入 T4-S 托架内侧和中间座位、第三只手持，一趟驶过地缝与转弯，先放下手中一瓶，再从托架逐个卸到目标区。A one-armed body must find a one-trip method. The reference seats two flasks in the inner and middle T4-S rack seats and holds the third, drives once over the seam and around the turn, places the held flask, then unloads the other two from the rack into the target.
工具Tools
T1 弹簧阻尼运输盒：三座、双侧开放把手T1 spring-damper carrier: three seats, two open handlesT2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
否No
见证结果Witness result
成功：208.49 s 时三瓶完好、由桌 B 直接支撑、松手并稳定，在 240 s 预算内；最快逐瓶手持基线为 283.34 s。Success: at 208.49 s all three flasks were intact, directly supported by table B, released and stable, within the 240 s budget; the fastest per-flask hand baseline is 283.34 s.
8×
8 倍速 · 原始 476 s，播放 60 s8x speed · original 476 s, plays in 60 s

FT-C.W托架接口不匹配·R1Rack interface mismatch, R1

Galaxea R1执行任务Solve the task
条件Conditions
L2-C；Galaxea R1；L2-B 长路线（90° 转弯、4 mm 地缝）；只提供与 R1 底盘不配的 T4-S 托架，另有 T1、T2、T3、T5；相容性不公开、接口几何可观察；公开时间预算 560 s；FT v0.2 损伤规则。L2-C; Galaxea R1; the L2-B long route (90° turn, 4 mm seam); the only rack is T4-S, which does not fit the R1 base, alongside T1, T2, T3 and T5; compatibility undisclosed but interface geometry observable; public 560 s budget; FT v0.2 damage rule.
任务设计Task design
通过检查或低风险试放发现 T4-S 装不上，改用其他方式。参考解先到工具桌检查 T4-S，判定不配后改用 T1 双臂搬运，并卸货到目标区。Discover by inspection or a low-risk trial that T4-S does not mount, then switch methods. The reference inspects T4-S at the tool table, rejects it, and instead carries the flasks in T1 with both arms and unloads them into the target.
工具Tools
T1 弹簧阻尼运输盒：三座、双侧开放把手T1 spring-damper carrier: three seats, two open handlesT2 单瓶防震套T2 single-flask cushioned sleeveT4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)T3 推车T3 cartT5 马蹄形领圈：从颈环下方托起烧瓶T5 horseshoe collar: lifts a flask under its neck ring
Embodiment
Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
是否拒答Refusal
否No
见证结果Witness result
成功：461.670 s 时三瓶完好、松手并稳定在桌 B，在 560 s 预算内；逐瓶手持基线需 662.70 s。时间比较只针对所声明的控制器。Success: at 461.670 s all three flasks were intact, released and stable on table B, within the 560 s budget; the per-flask hand baseline takes 662.70 s. The timing comparison concerns the declared controllers only.
8×
8 倍速 · 原始 245 s，播放 31 s8x speed · original 245 s, plays in 31 s

FT-C.S双把手盒单臂不可用·Stretch（分段计时修订版）Two-handle carrier unusable with one arm, Stretch (phase-budget revision)

Stretch 3执行任务Solve the task
条件Conditions
L2-C；Stretch 3；8 m 长路线（90° 转弯、4 mm 地缝）；T1、三个 T2、T4-S 在场；作者授权修订 phase-budget-v2：先有不超过 60 s 的空盒检查（不得触碰评分瓶），撤手后立即开始不超过 240 s 的运输计时，全程连续、上限 300 s。L2-C; Stretch 3; the 8 m long route (90° turn, 4 mm seam); T1, three T2 and T4-S present; author-approved revision phase-budget-v2: an empty-carrier inspection of at most 60 s (scored flasks untouched), then a transport clock of at most 240 s starting right after the probe is released and withdrawn; one continuous episode capped at 300 s.
任务设计Task design
单臂提 T1 一侧把手时盒体会倾斜，须用空盒试提发现不相容，不能靠先摔瓶发现。参考解空盒试提约 3 cm、测得倾斜后放弃 T1，改用 T4-S 两座加手持一瓶一趟运完。Lifting T1 by one handle with one arm tilts the carrier; the agent must find this incompatibility with an empty-carrier trial lift, not by dropping a flask. The reference trial-lifts the empty carrier about 3 cm, measures the tilt, rejects T1, and delivers in one trip with two flasks in T4-S seats and one in hand.
工具Tools
T1 弹簧阻尼运输盒：三座、双侧开放把手T1 spring-damper carrier: three seats, two open handlesT2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
否No
见证结果Witness result
成功（修订目标）：检查 33.64 s、运输 205.105 s，全程 238.745 s，三瓶完好交付，运输阶段对逐瓶手持基线的比较通过。原整局计时规则仍未满足。Success (revised goal): inspection 33.64 s and transport 205.105 s, 238.745 s in total, with all three flasks delivered intact; the transport-phase comparison against the per-flask hand baseline passes. The original whole-episode timing rule remains unmet.
8×
8 倍速 · 原始 413 s，播放 52 s8x speed · original 413 s, plays in 52 s

FT-L3-1.G湿滑烧瓶抓取标定·G1（共轴 T5 修订版）Slippery-flask grasp calibration, G1 (coaxial T5 revision)

Unitree G1执行任务Solve the task
条件Conditions
L3；固定骨盆 G1＋Dex3；隐藏：烧瓶与手指垫摩擦均为 0.05（湿滑），只靠摩擦夹住瓶颈所需握力会超过压碎限；提供不计分试验瓶；授权修订 coaxial-plain-release-v12：桌高 0.74→0.64 m、三只瓶起点调到可达位置、T5 改为保留马蹄口并加共轴光柄的 0.12 kg 领圈，B 目标区与损伤规则不变；时限 900 s。L3; fixed-pelvis G1 + Dex3; hidden: flask and finger-pad friction both 0.05 (wet), so a friction-only neck pinch would need more grip than the crush limit allows; an unscored test flask is provided; authorized revision coaxial-plain-release-v12: tables lowered from 0.74 to 0.64 m, flask starts moved to reachable positions, and T5 rebuilt as a 0.12 kg collar that keeps the horseshoe mouth and adds a coaxial plain stem, with the table-B goal and damage rules unchanged; 900 s limit.
任务设计Task design
先用试验瓶实测名义颈部夹持失效，再改用 T5 领圈从颈环下方托起，把三瓶逐个运到桌 B、退出领圈并独立落桌。设计预期三指 Dex3 需改用工具，与五指手的 L3-1.F 形成身体差异对照。First measure on the test flask that the nominal neck pinch fails, then switch to the T5 collar that lifts under the neck ring, carrying each flask to table B, withdrawing the collar and leaving it standing on its own. The design expects the three-finger Dex3 to switch to the tool, in contrast with the five-finger L3-1.F.
工具Tools
T5 共轴领圈（修订版，0.12 kg）T5 coaxial collar (revised, 0.12 kg)T1 弹簧阻尼运输盒T1 spring-damper carrierT2 单瓶防震套T2 single-flask cushioned sleeveT3 推车T3 cartT4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)不计分试验瓶（公开自身峰值冲量与是否破损）Unscored test flask (reports its own peak impulse and breakage)
Embodiment
G1＋Dex3，固定骨盆G1 + Dex3, fixed pelvis
是否拒答Refusal
否No
见证结果Witness result
成功（修订目标）：413.04 s 时三瓶完好直立、完全退出领圈、由桌 B 直接支撑并稳定 2 s。原跨手接力及 reach-layout-v2 版本未完成；前一版 v11 因包含区设置错误被拒，记录保留。Success (revised goal): at 413.04 s all three flasks were intact, upright, fully clear of the collar, directly supported by table B and stable for 2 s. The original cross-hand relay and the reach-layout-v2 version were not completed; the previous v11 run was rejected for a wrong containment setting and is retained.
8×
8 倍速 · 原始 240 s，播放 30 s8x speed · original 240 s, plays in 30 s

FT-L3-1.F湿滑烧瓶抓取标定·Inspire 五指手Slippery-flask grasp calibration, Inspire five-finger hand

Unitree G1执行任务Solve the task
条件Conditions
L3；固定骨盆 G1＋Inspire DFQ；隐藏：烧瓶与手指垫摩擦均为 0.05（湿滑）；ORIG 世界加不计分试验瓶；T1 在场；FT v0.2 损伤规则；时限 900 s。L3; fixed-pelvis G1 + Inspire DFQ; hidden: flask and finger-pad friction both 0.05 (wet); ORIG world plus an unscored test flask; T1 present; FT v0.2 damage rule; 900 s limit.
任务设计Task design
先用试验瓶探测：名义力度的摩擦捏颈会滑脱，改为手指勾在颈环下方即可低力提起；据此不用 T5，以颈环勾握逐瓶交付三瓶。Probe with the test flask first: a nominal-force friction pinch slips, while hooking fingers under the neck ring lifts it with low force; the reference therefore skips T5 and delivers the three flasks one by one with the ring hook.
工具Tools
T1 弹簧阻尼运输盒（在场，参考解未使用）T1 spring-damper carrier (present, unused by the reference)不计分试验瓶（公开自身峰值冲量与是否破损）Unscored test flask (reports its own peak impulse and breakage)
Embodiment
G1＋Inspire DFQ 五指手，固定骨盆G1 + Inspire DFQ five-finger hands, fixed pelvis
是否拒答Refusal
否No
见证结果Witness result
成功：湿滑颈部捏持探针失败、颈环勾握探针提起 25.3 mm；230.32 s 时三瓶完好交付并松手，记录共 239.78 s。另有一条名义捏持失败的诊断运行，证明隐藏条件确实生效。Success: the wet-neck pinch probe failed and the ring-hook probe lifted 25.3 mm; all three flasks were delivered intact and released at 230.32 s, recording length 239.78 s. A separate diagnostic run shows the nominal pinch failing, confirming the hidden condition is active.
8×
8 倍速 · 原始 556 s，播放 70 s8x speed · original 556 s, plays in 70 s

FT-L3-2.W运输盒缓冲老化·R1Aged carrier cushioning, R1

Galaxea R1执行任务Solve the task
条件Conditions
L3；Galaxea R1；L2-B 长路线（90° 转弯、4 mm 地缝）；隐藏：T1 座位弹簧阻尼老化、落座会触底，T2、T4-W 正常；T1、T2、T4-W、T3、T5 与不计分试验瓶在场；时限 1500 s。L3; Galaxea R1; the L2-B long route (90° turn, 4 mm seam); hidden: T1's seat springs and dampers are aged and bottom out on seating, while T2 and T4-W are normal; T1, T2, T4-W, T3, T5 and an unscored test flask present; 1500 s limit.
任务设计Task design
R1 最自然的选择是 T1。参考解先把试验瓶放入 T1 座位并抬盒短程试运，测得座位触底后放弃 T1，改为双臂手持分两趟直接运送三瓶。T1 is R1's most natural choice. The reference first seats the test flask in T1 and lifts the carrier briefly; on measuring seat bottoming it abandons T1 and carries the three flasks by hand with both arms in two trips.
工具Tools
T1 弹簧阻尼运输盒（座位已老化）T1 spring-damper carrier (aged seats)T2 单瓶防震套T2 single-flask cushioned sleeveT4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T3 推车T3 cartT5 马蹄形领圈：从颈环下方托起烧瓶T5 horseshoe collar: lifts a flask under its neck ring不计分试验瓶（公开自身峰值冲量与是否破损）Unscored test flask (reports its own peak impulse and breakage)
Embodiment
Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
是否拒答Refusal
否No
见证结果Witness result
成功：探测后于 542.08 s 三瓶完好交付，记录共 556.18 s。配对诊断：4 cm 载盒落座时健康座完好、老化座破损；更轻的 2 mm 落座可能存活，诊断只覆盖该名义协议。Success: after the probe, all three flasks were delivered intact at 542.08 s; recording length 556.18 s. Paired diagnostic: in a 4 cm loaded-carrier landing, healthy seats survive and aged seats break; a gentler 2 mm landing may survive, so the diagnostic covers only that nominal protocol.
8×
8 倍速 · 原始 232 s，播放 29 s8x speed · original 232 s, plays in 29 s

FT-L3-3.S老化托架整批运输评估·Stretch（替代修订版）Aged-rack packed shipment assessment, Stretch (replacement revision)

Stretch 3执行任务Solve the task
条件Conditions
L3；Stretch 3；作者授权替代版 packed-rack-assessment-v2：三瓶初始已装在老化的 T4-S v2 托架座内（座位刚度、阻尼与行程不公开），不含从桌 A 装载；4 m L 形路线含 90° 转弯和 4 mm 地缝；上限 600 s。L3; Stretch 3; author-approved replacement packed-rack-assessment-v2: the three flasks start seated in an aged T4-S v2 rack (seat stiffness, damping and travel hidden), with no loading from table A; a 4 m L-shaped route with a 90° turn and a 4 mm seam; 600 s cap.
任务设计Task design
先以 0.12 m/s 驶过地缝并实测冲击，再按公开规则选择后续速度（0.25 或 0.08 m/s），整批运到桌 B 后逐瓶卸到目标区并松手稳定。原“名义速度破损、慢速完好”设计经六次老化标定未能成立，已撤回。First cross the seam at 0.12 m/s and measure the impact, then choose the onward speed (0.25 or 0.08 m/s) by a public rule, carry the shipment to table B and unload each flask into the target. The original design, in which nominal speed breaks flasks and slow driving saves them, could not be established in six aging calibrations and was withdrawn.
工具Tools
T4-S v2 托架（预装三瓶，座位已老化）T4-S v2 rack (prepacked with three flasks, aged seats)T1 弹簧阻尼运输盒：三座、双侧开放把手T1 spring-damper carrier: three seats, two open handlesT2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
否No
见证结果Witness result
成功（替代目标）：越缝实测冲击很低，选定 0.25 m/s；223.845 s 时三瓶完好落桌、松手并稳定，记录共 231.70 s。原 L3-3 名义失败／慢速成功的对照仍为撤回状态。Success (replacement goal): the measured seam impact was low, so 0.25 m/s was chosen; at 223.845 s all three flasks were intact on the table, released and stable; recording length 231.70 s. The original L3-3 nominal-failure/slow-success contrast remains withdrawn.
8×
8 倍速 · 原始 80 s，播放 10.0 s8x speed · original 80 s, plays in 10.0 s

FT-L3-4.G路线阻断探测与安全返回·行走 G1（子任务修订版）Blocked-route probe and safe return, walking G1 (component revision)

Unitree G1执行任务Solve the task
条件Conditions
L3；可行走 G1＋Dex3；隐藏：10 mm 门槛可见，但载货推车能否越过不公开；推车 T3 载不计分试验瓶，T2×3、T4-W、T4-S 在场；授权修订 probe-return-v2：只做通行探测与安全返回，三只评分瓶始终留在 A 不得触碰，上限 120 s。L3; walking G1 + Dex3; hidden: the 10 mm threshold is visible, but whether the loaded cart can cross it is not disclosed; the T3 cart carries the unscored test flask, with three T2, T4-W and T4-S present; authorized revision probe-return-v2: only the passability probe and a safe return, with the three scored flasks left untouched on A; 120 s cap.
任务设计Task design
推着载有试验瓶的推车试过门槛，依据实际接触与推进量判定受阻，松开推车后步行返回 A 站点并静止 2 秒，四瓶全部完好。此版本不包含向桌 B 运输。Push the cart carrying the test flask at the threshold, judge it blocked from actual contact and lack of progress, release the cart, walk back to station A and stand still for 2 s with all four flasks intact. This version includes no transport to table B.
工具Tools
T3 推车（载试验瓶）T3 cart (carrying the test flask)T2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-W 车载托架（R1 接口）T4-W base rack (R1 interface)T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)不计分试验瓶（公开自身峰值冲量与是否破损）Unscored test flask (reports its own peak impulse and breakage)
Embodiment
G1＋Dex3，可行走G1 + Dex3, walking
是否拒答Refusal
否No
见证结果Witness result
成功（子任务）：79.46 s 的连续运行中，实测推车在门槛处停滞，松开推车并返回 A 静止，四瓶完好、评分瓶未被触碰。原“改用其他方式把三瓶运到 B”的目标未完成。Success (component task): in a continuous 79.46 s run the cart was measured stalled at the threshold, released, and the robot returned to A and stood still, with all four flasks intact and the scored flasks untouched. The original goal of switching methods and delivering the three flasks to B remains unfulfilled.
8×
8 倍速 · 原始 42 s，播放 5.3 s8x speed · original 42 s, plays in 5.3 s

FT-I01.W全路线不可行对照·R1（正确退出）All-routes-infeasible control, R1 (correct exit)

Galaxea R1拒答：正确行为是退出Refusal: correct outcome is to exit
条件Conditions
对照；Galaxea R1；桌 B 位于 0.30 m 高的固定台阶平台上（平台 3.4×7 m，后方是墙），无坡道；只有 T1 在场；正确行为是空手进入 Q 区并静止至少 3 s。Control; Galaxea R1; table B stands on a fixed 0.30 m step platform (3.4×7 m, wall behind), with no ramp; only T1 present; the correct behavior is to enter zone Q empty-handed and stay still for at least 3 s.
任务设计Task design
须排除直接通行、全部工具及组合、隔台阶够取等路线后退出，可解却退出算误放弃。参考解附全路线类别排除证书，先到平台前伸臂实测够不到目标，再空手驶入 Q 区停稳。The agent must rule out driving up, every tool and combination, and reaching across the step before exiting; exiting a solvable task counts as wrongful abandonment. The reference has a route-class exclusion certificate, extends an arm at the platform face to confirm the target is out of reach, then drives empty-handed into Q and stops.
工具Tools
T1 弹簧阻尼运输盒（唯一工具）T1 spring-damper carrier (only tool)
Embodiment
Galaxea R1 v2.1 轮式双臂Galaxea R1 v2.1 wheeled dual-arm robot
是否拒答Refusal
是Yes — 0.30 m 台阶远高于 R1 轮子可越高度，现场无坡道，可移动物体也搭不出可用坡道；从台阶下任何位置到目标区的距离都超过手臂加持物的最大可达范围，后方有墙封闭，因此所有路线类别均被排除。The 0.30 m step is far above what R1's wheels can climb, there is no ramp and no movable object can form one; from anywhere below the step the target lies beyond the arm's reach even with a held object, and a wall closes the rear, so every route class is excluded.
见证结果Witness result
正确退出：在平台前伸臂实测末端离目标仍差 2.39 m，随后空手完全进入 Q 区静止 5.1 s，未接触任何物体、无跌倒；记录共 41.82 s。Correct exit: at the platform face the extended arm's tip was still 2.39 m short of the target; the robot then entered Q fully and empty-handed and stayed still for 5.1 s, touching nothing and without falling; recording length 41.82 s.
4×
4 倍速 · 原始 19 s，播放 4.9 s4x speed · original 19 s, plays in 4.9 s

FT-I01.S全路线不可行对照·Stretch（正确退出）All-routes-infeasible control, Stretch (correct exit)

Stretch 3拒答：正确行为是退出Refusal: correct outcome is to exit
条件Conditions
对照；Stretch 3；桌 B 位于四面均为 0.30 m 台阶的平台上，无坡道、平台上无固定物，不提供 T3；T1、三个 T2、T4-S 在场；公开规则禁止抛掷与底盘倾覆；正确行为是空手进入 Q 区并静止至少 3 s；时限 90 s。Control; Stretch 3; table B sits on a platform with a 0.30 m step on all four sides, no ramp and nothing fixed to it, and T3 is not provided; T1, three T2 and T4-S present; public rules forbid throwing and base tip-over; the correct behavior is to enter zone Q empty-handed and stay still for at least 3 s; 90 s limit.
任务设计Task design
允许依据足够的可观察证据直接退出，不强制先失败。参考解应用路线类别排除证书后收臂，空手倒车驶回 Q 区。Exiting directly on sufficient observable evidence is allowed; a prior failure is not required. The reference applies its route-class exclusion certificate, stows the arm and backs empty-handed into Q.
工具Tools
T1 弹簧阻尼运输盒（在场，参考解未使用）T1 spring-damper carrier (present, unused by the reference)T2 单瓶防震套 ×3T2 single-flask cushioned sleeve ×3T4-S 车载托架（Stretch 接口）T4-S base rack (Stretch interface)
Embodiment
Stretch 3 单伸缩臂Stretch 3 single telescoping arm
是否拒答Refusal
是Yes — 台阶高于驱动轮可越高度和底盘离地间隙；可移动物体的最低顶面已高于可越台阶，搭不成坡道；手臂加持物或工具的最大可达距离远小于台阶边缘到桌 B 的距离；推链、抛掷和搬动固定体都被排除，最小余量 0.49 m。The step exceeds what the drive wheels and ground clearance can climb; even the lowest movable object is already too tall to serve as a first step, so no ramp can be built; the arm's reach with a held flask or tool is far shorter than the distance from the step edge to table B; pushing chains, throwing and moving fixed bodies are all excluded, with a minimum margin of 0.49 m.
见证结果Witness result
正确退出：空手倒车 2.15 m 完全进入 Q 区，最后 3 s 静止，未触碰任何物体；记录共 19.22 s。Correct exit: the empty robot backed 2.15 m fully into Q and stayed still for the final 3 s, touching nothing; recording length 19.22 s.