五维平均 Score
20.84参考位置 #6
五维平均成功率
14.40%按 SR 参考位置 #7
原始回合成功率
15.16%955 / 6,300 成功
Focus 任务
26 / 26全部测完 · 2,775 回合
五个能力维度
维度内任务等权,五维等权;泛化维度的普通与随机布局各占一半。Score 包含部分完成得分,SR 仅统计完整成功。
| 维度 | 75k Score | 85k Score | 85k − 75k |
|---|---|---|---|
| 泛化 | 24.94 | 25.68 | +0.74 |
| 精度 | 33.91 | 33.08 | -0.83 |
| 长程 | 34.04 | 32.16 | -1.88 |
| 记忆 | 6.81 | 5.43 | -1.37 |
| 开放指令 | 4.49 | 4.10 | -0.39 |
54 个任务配置
Δ 为 85k − 75k 在共同 (task, seed, layout) 上的成功率变化,单位为百分点。展开“逐回合”可查看每个布局的结果;绿色表示成功,悬停或聚焦布局编号可查看部分得分。
54 / 54 个任务配置
| 任务 / 维度 | 完成回合 | 成功数 / SR | Score | 共同布局 | Δ SR / pp | 详情 |
|---|---|---|---|---|---|---|
| stack_bowlsfocus 泛化 · 新增训练轨迹 174 | 75/75 | 63 · 84.0% | 86.00 | 75 | +1.33 | 逐回合Stack the three bowls together. 加载中… |
| put_bottles_into_dustbinfocus 长程 · 新增训练轨迹 173 | 150/150 | 139 · 92.7% | 95.40 | 150 | +2.00 | 逐回合Pick up the bottles and throw them into the dustbin, using handover when needed. 加载中… |
| build_towerfocus 精度 · 新增训练轨迹 107 | 150/150 | 90 · 60.0% | 67.40 | 150 | +0.00 | 逐回合Build a tower using the wooden blocks and wooden boards. 加载中… |
| stack_blocksfocus 泛化 · 新增训练轨迹 100 | 75/75 | 49 · 65.3% | 70.13 | 75 | -4.00 | 逐回合Stack the three blocks with different textures. 加载中… |
| insert_tubesfocus 精度 · 新增训练轨迹 96 | 150/150 | 82 · 54.7% | 69.20 | 150 | -14.67 | 逐回合Insert the three tubes into the rack one by one. 加载中… |
| fold_clothesfocus 泛化 · 新增训练轨迹 94 | 75/75 | 32 · 42.7% | 51.73 | 75 | -5.33 | 逐回合Fold the clothes neatly. 加载中… |
| make_kongfocus 长程 · 新增训练轨迹 88 | 150/150 | 25 · 16.7% | 16.67 | 150 | -4.67 | 逐回合Wait for the opponent to discard a tile, then declare a kong with the matching tiles. 加载中… |
| pour_balls_into_vasefocus 精度 · 新增训练轨迹 83 | 150/150 | 56 · 37.3% | 37.33 | 150 | +8.00 | 逐回合Pour all the balls from the cup into the vase. 加载中… |
| pour_liquid_into_cupfocus 泛化 · 新增训练轨迹 77 | 75/75 | 29 · 38.7% | 38.67 | 75 | +6.67 | 逐回合Pour the liquid from the bottle into the cup. 加载中… |
| fill_pen_holderfocus 长程 · 新增训练轨迹 66 | 150/150 | 41 · 27.3% | 48.67 | 150 | -2.00 | 逐回合Hold the pen holder with one hand, place all pens into it with the other hand, then put it back down. 加载中… |
| make_toastfocus 泛化 · 新增训练轨迹 65 | 75/75 | 33 · 44.0% | 55.33 | 74 | +10.81 | 逐回合Pick up two slices of bread, place them into the toaster, and press the lever down. 加载中… |
| store_laptop_and_headphonesfocus 泛化 · 新增训练轨迹 48 | 75/75 | 14 · 18.7% | 35.47 | 73 | +0.00 | 逐回合Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand. 加载中… |
| hang_mugsfocus 泛化 · 新增训练轨迹 46 | 75/75 | 25 · 33.3% | 51.80 | 75 | -6.67 | 逐回合Hang all the mugs on the mug rack. 加载中… |
| fold_clothes_randomfocus 泛化 · 新增训练轨迹 38 | 75/75 | 15 · 20.0% | 28.00 | 75 | -14.67 | 逐回合Fold the clothes neatly. 加载中… |
| fasten_screwsfocus 精度 · 新增训练轨迹 33 | 150/150 | 33 · 22.0% | 52.47 | 150 | +0.00 | 逐回合Insert and tighten each screw into the nut of the same color. 加载中… |
| plug_in_chargerfocus 精度 · 新增训练轨迹 33 | 150/150 | 15 · 10.0% | 10.00 | 150 | -4.67 | 逐回合Plug the charger into the power strip. 加载中… |
| pour_liquid_into_cup_randomfocus 泛化 · 新增训练轨迹 32 | 75/75 | 24 · 32.0% | 32.00 | 75 | -1.33 | 逐回合Pour the liquid from the bottle into the cup. 加载中… |
| general_pickupfocus 开放指令 · 新增训练轨迹 31 | 150/150 | 27 · 18.0% | 18.00 | 150 | -1.33 | 逐回合Pick up the <target> by 10 cm. 加载中… |
| pack_objects_into_boxfocus 泛化 · 新增训练轨迹 30 | 75/75 | 11 · 14.7% | 39.80 | 75 | +8.00 | 逐回合Place all the objects into the box with their front sides facing left. 加载中… |
| classify_objectsfocus 长程 · 新增训练轨迹 14 | 150/150 | 8 · 5.3% | 12.37 | 149 | +2.01 | 逐回合Sort the objects by category into the three baskets. 加载中… |
| sort_nesting_dolls_by_sizefocus 泛化 · 新增训练轨迹 3 | 75/75 | 13 · 17.3% | 17.33 | 73 | +4.11 | 逐回合Arrange the five nesting dolls in a row from left to right, from smallest to largest. 加载中… |
| make_toast_randomfocus 泛化 · 新增训练轨迹 2 | 75/75 | 6 · 8.0% | 16.67 | 73 | +1.37 | 逐回合Pick up two slices of bread, place them into the toaster, and press the lever down. 加载中… |
| stack_blocks_by_languagefocus 开放指令 · 新增训练轨迹 2 | 150/150 | 21 · 14.0% | 16.13 | 150 | -3.33 | 逐回合stack the blocks from bottom to top in the order of {color[str(cat_index[0])]}, {color[str(cat_index[1])]}, and {color[str(cat_index[2])]}, then reset the robot arm. 加载中… |
| pack_objects_into_box_randomfocus 泛化 · 新增训练轨迹 1 | 75/75 | 2 · 2.7% | 22.87 | 73 | +0.00 | 逐回合Place all the objects into the box with their front sides facing left. 加载中… |
| stack_bowls_randomfocus 泛化 · 新增训练轨迹 1 | 75/75 | 1 · 1.3% | 13.53 | 75 | +2.67 | 逐回合Stack the three bowls together. 加载中… |
| store_laptop_and_headphones_randomfocus 泛化 · 新增训练轨迹 1 | 75/75 | 2 · 2.7% | 15.47 | 69 | +0.00 | 逐回合Hang the headphones on the headphone stand, close the laptop, then place it into the vertical laptop stand. 加载中… |
| align_blocks 开放指令 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Use the set square to push the three blocks into a straight, aligned row, then reset the robot arm. 加载中… |
| arrange_largest_number 泛化 | 75/75 | 0 · 0.0% | 7.27 | 75 | +1.33 | 逐回合Arrange the numbers from left to right to form the largest possible number, and place them on the pad. 加载中… |
| arrange_largest_number_random 泛化 | 75/75 | 0 · 0.0% | 2.27 | 75 | +0.00 | 逐回合Arrange the numbers from left to right to form the largest possible number, and place them on the pad. 加载中… |
| classify_objects_by_language 开放指令 | 150/150 | 0 · 0.0% | 1.27 | 150 | +0.67 | 逐回合Put {cat0} objects into the left basket, {cat1} objects into the middle basket, and {cat2} objects into the right basket, then reset the robot arm. 加载中… |
| cover_blocks 记忆 | 150/150 | 10 · 6.7% | 14.00 | 150 | +2.00 | 逐回合Cover the blocks from left to right, remember their colors, then uncover them in the order: red, green, and blue. 加载中… |
| deposit_coin 精度 | 150/150 | 16 · 10.7% | 18.00 | 150 | +3.33 | 逐回合Pick up the coin from the holder and insert it precisely into the coin bank. 加载中… |
| fill_egg_holder 长程 | 150/150 | 1 · 0.7% | 8.40 | 150 | -0.67 | 逐回合Place the four eggs from the basket into the egg holder, then close the lid. 加载中… |
| hang_mugs_random 泛化 | 75/75 | 0 · 0.0% | 3.53 | 75 | +0.00 | 逐回合Hang all the mugs on the mug rack. 加载中… |
| imitate_sorting_sequence 记忆 | 150/150 | 0 · 0.0% | 1.50 | 146 | +0.68 | 逐回合Observe the object placement order, remember it, then place the corresponding objects into the basket in the same order. 加载中… |
| insert_key 精度 | 150/150 | 0 · 0.0% | 14.20 | 150 | +0.00 | 逐回合Pick up the key, hand it over to the other hand, insert it into the keyhole, then turn it. 加载中… |
| match_and_pick_from_conveyor 记忆 | 150/150 | 38 · 25.3% | 25.33 | 150 | -12.00 | 逐回合Remember the first object on the conveyor, then pick the matching object when it appears again. 加载中… |
| organize_table 长程 | 150/150 | 19 · 12.7% | 53.50 | 150 | -4.00 | 逐回合Place the alarm clock on the drawer, put the figurine on the stand, place the mouse on the mouse pad, and push the keyboard into the frame. 加载中… |
| pick_from_conveyor_by_image 开放指令 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Lift the basket more than 8 cm, identify the target object on the conveyor according to the image on the board, pick it up, and place it into the basket. 加载中… |
| play_Xylophone 精度 | 150/150 | 4 · 2.7% | 2.67 | 150 | -1.33 | 逐回合Pick up the mallet and strike all xylophone keys from left to right. 加载中… |
| play_stacking_toy 长程 | 150/150 | 0 · 0.0% | 4.40 | 150 | +0.00 | 逐回合Place all stacking toy pieces onto the correct pegs. 加载中… |
| play_tic_tac_toe 长程 | 150/150 | 9 · 6.0% | 32.93 | 143 | -0.70 | 逐回合Play tic-tac-toe as the first player and fill the board with the opponent. 加载中… |
| pour_by_language 开放指令 | 150/150 | 0 · 0.0% | 0.00 | 148 | +0.00 | 逐回合bottle_0 加载中… |
| press_by_number 记忆 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Press the two red buttons the required number of times according to the number cards, then press the blue button to confirm. 加载中… |
| push_T 泛化 | 75/75 | 0 · 0.0% | 0.00 | 69 | +0.00 | 逐回合Push the T-shaped block to align it precisely with the gray T-shaped pad. 加载中… |
| push_T_random 泛化 | 75/75 | 0 · 0.0% | 0.00 | 75 | +0.00 | 逐回合Push the T-shaped block to align it precisely with the gray T-shaped pad. 加载中… |
| solve_equation 开放指令 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Complete the equation by selecting the correct missing number or operator and placing it on the pad, then reset the robot arm. 加载中… |
| sort_nesting_dolls_by_size_random 泛化 | 75/75 | 0 · 0.0% | 0.00 | 75 | +0.00 | 逐回合Arrange the five nesting dolls in a row from left to right, from smallest to largest. 加载中… |
| stack_blocks_random 泛化 | 75/75 | 0 · 0.0% | 8.00 | 75 | +0.00 | 逐回合Stack the three blocks with different textures. 加载中… |
| store_tools_in_toolbox 开放指令 | 150/150 | 0 · 0.0% | 0.50 | 150 | +0.00 | 逐回合Place each tool into its matching position in the toolbox, then reset the robot arm. 加载中… |
| swap_T 记忆 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Pick up the two T-shaped blocks, swap their positions, and place them back with the correct orientations. 加载中… |
| swap_blocks 记忆 | 150/150 | 0 · 0.0% | 0.00 | 150 | +0.00 | 逐回合Swap the two blocks using the empty mat, pressing the button after each move. 加载中… |
| sweep_blocks 泛化 | 75/75 | 2 · 2.7% | 2.67 | 75 | -1.33 | 逐回合Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the blocks. 加载中… |
| sweep_blocks_random 泛化 | 75/75 | 0 · 0.0% | 0.00 | 73 | +0.00 | 逐回合Pick up the broom, hand it over to the right hand, then use the dustpan to sweep the objects. 加载中… |
与官网榜单对照
官网 2026.9.23 版的 48 个模型,加上本页 checkpoint,按 Score 排序。每份 checkpoint 分别插入作参考;这不是正式上榜。下方显示前 10。
| 参考位置 | 模型 | Score | 五维 SR |
|---|---|---|---|
| #1 | Simate-beta | 33.95 | 27.96% |
| #2 | Liber-0 Preview | 30.74 | 25.52% |
| #3 | Liber-0 Lite | 29.24 | 24.23% |
| #4 | GPT-6-Astra | 28.97 | 22.48% |
| #5 | DM0.5 | 24.90 | 19.34% |
| #6 | DO-WAM 75k EMA | 20.84 | 14.40% |
| #7 | GalaxeaVLA (G0.5) | 20.23 | 14.88% |
| #8 | Xiaomi-Robotics-1 | 20.07 | 13.93% |
| #9 | OpenWAM-α | 17.18 | 11.92% |
| #10 | Meituan-Robotics-0 | 14.95 | 9.53% |
评测口径与数据说明
- 纯 WAM;官方 RoboDojo Docker、CPU 物理、单环境;denoise 10 / replan 32;seed 0/1/2。42 个基础任务含随机化变体共 54 个配置,计划 6,300 回合。
- 75k 为 WSD-S 第 2 周期 checkpoint,85k 为在其后继续训练 10,000 步的第 3 周期 checkpoint。Focus 为新增 rollout 实际参与训练的 26 个任务配置。
- 已知训练数据与官方评测有 63 个布局重合。剔除已记录重合后,本页五维 Score 为 20.65,SR 为 14.20%。这项复核不代表测试集已完全独立。
- 60k 每任务只有 21 次;与 75k / 85k 的比较见 Focus 配对报告,使用三方共同布局。
- 此页为发布时快照,数据时间:2026-09-26 20:27:24 UTC。Run:cfgexp75k_wsds10k_c02_official_docker_20260926。