Both versions are aligned block by block, in reading order: title, the essentials, then paragraph by paragraph. Where the translation merged or split a paragraph, the matching cell stays empty — we never pair two passages by guesswork.
传送带不停地向前送包裹：纸箱、软袋、圆柱形快件和泡沫封装的生鲜件，大小、重量、材质和朝向都没有统一规则。据这场公开直播演示，自变量的双臂机器人只装着工业中常见的标准夹爪，却在连续一小时直播中完成了1816件有效分拣。按短时吞吐量计算，这一成绩比Figure 03此前连续运行200小时留下的1248件/小时标杆高出45%。
The conveyor keeps moving packages forward: cartons, soft bags, cylindrical parcels and foam-packed fresh-food items, with no uniform rules governing their size, weight, material or orientation. According to this public livestream demonstration, X Square Robot's dual-arm robot was fitted only with standard grippers commonly used in industry, yet completed 1,816 valid sorts during one continuous hour of livestreaming. Calculated by short-term throughput, this result was 45% higher than the 1,248 packages/hour benchmark that Figure 03 had set during 200 consecutive hours of operation.
两家公司采用了不同的硬件配置。Figure 03保留完整人形身体和五指灵巧手；自变量则去掉分拣场景用不到的腿和手指，把任务交给两条机械臂和夹爪完成。公开信息称，自变量方案的成本下降70%，分拣准确率超过98%。它不靠固定程序处理每一种包裹，而是先识别大小、形状、材质和姿态，再决定抓取力度与移动路径。
The two companies use different hardware configurations. Figure 03 retains a full humanoid body and five-finger dexterous hands; X Square Robot, by contrast, removed the legs and fingers unnecessary for sorting and assigned the task to two robotic arms and grippers. Public information says the X Square Robot solution cuts costs by 70%, with a sorting accuracy rate above 98%. Rather than relying on fixed programs for each type of package, it first identifies size, shape, material and posture, then determines the gripping force and movement path.
这套判断由WALL-B完成。自变量于2026年4月发布这一具身智能大模型，并采用WUM架构，也就是把视觉、语言、动作以及对物理变化的预测放进同一个网络联合训练。面对轻薄快递袋，机器人可以用一只夹爪快速抓取；遇到大纸箱，则改用双臂托举或从侧面推动。需要翻转面单时，轻件借助惯性快速改变姿态，重箱分步调整，容易变形的软包则先拨开、摊平，再校准方向。
This decision-making is handled by WALL-B. X Square Robot released this embodied-intelligence foundation model in April 2026 and uses the WUM architecture, which jointly trains vision, language, actions and predictions of physical changes in a single network. When faced with a thin delivery bag, the robot can quickly grab it with one gripper; for a large carton, it switches to supporting it with both arms or pushing it from the side. When a shipping label needs to be turned over, a light item can use inertia to change orientation quickly, a heavy box can be adjusted in stages, and a soft package that is prone to deformation can first be nudged aside and flattened before its direction is recalibrated.
这条路线并不意味着硬件不再重要。夹爪开口、摩擦力、机械臂负载、摄像头位置和力控精度，仍然决定机器人能处理哪些物体。WALL-B改变的是硬件的使用方式：原本只能执行单一动作的夹爪，可以根据包裹状态组合出夹、推、拨、摊和双臂托举等策略。自变量也曾在2025年发布单手20个自由度的五指灵巧手，这次使用标准夹爪，更多是因为仓库工位不需要为行走、平衡和十根手指付费。
This approach does not mean hardware is no longer important. Gripper opening, friction, robotic-arm load, camera position and force-control precision still determine which objects the robot can handle. What WALL-B changes is how the hardware is used: a gripper that could previously perform only a single action can combine strategies such as gripping, pushing, nudging, flattening and supporting with both arms based on the package's condition. X Square Robot also released a five-finger dexterous hand with 20 degrees of freedom in 2025; its use of standard grippers this time is mainly because warehouse workstations do not need to pay for walking, balancing and ten fingers.
具体来说，企业比较的是两种面向不同工位的机器人配置。如果模型能力能够从家庭场景复用到仓库，再从一个工位复制到更多工位，部署工作就可能更多集中在机械适配和流程调整，而不是每条产线重新采集数据、设计动作和长期调试。当前演示已经展示了吞吐量、准确率和成本方向，但要进入规模采购，还必须继续验证运行稳定性、维护费用和回本周期。
Specifically, the companies are comparing two robot configurations designed for different workstations. If model capabilities can be reused from household settings in warehouses, and then replicated from one workstation to more workstations, deployment could focus more on mechanical adaptation and process adjustments rather than collecting data, designing movements and carrying out long-term tuning again for every production line. The current demonstration has shown the direction of throughput, accuracy and cost, but entering large-scale procurement will still require continued validation of operating stability, maintenance expenses and payback periods.