# X Square WALL-B 分拣效率超 Figure 45%

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-08-14 20:57
- AIHOT 分数：39
- AIHOT 链接：https://aihot.virxact.com/items/cmssz8qkf0n0nroffdxu936d5
- 原文链接：https://x.com/rohanpaul_ai/status/2088248739916099791

## AI 摘要

X Square 的 WALL-B 机器人直播分拣 1,816 件/小时，准确率超 98%，比 Figure 机器人 5 月耐力测试的 1,248 件/小时高出约 45%。其采用双臂与专用夹爪而非人形本体，连续重规划应对包裹遮挡、变形等变化。该模型正拓展至家庭场景，但泛化能力仍需验证。

## 正文

1,816 parcels/hour is fast.

For context, Figure Robot's May endurance run averaged 1,248/hour across 200 hours.

So @XSquareRobot 's 1-hour livestream comes out roughly 45% higher on throughput.

There is a pretty important architectural reason for that.

X Square isn't forcing the problem through a humanoid body.

WALL-B controls two High-Performance 6-Axis Robot Arms with specialized grippers, continuously replanning as parcels move, get occluded, fail to transfer, or need their barcode side reoriented.

The pile changes after every pick. Parcels overlap, soft bags deform, labels face the wrong way, transfers fail.

WALL-B is supposed to reassess that new state, choose another grasp or reorientation, and keep the station running instead of handing the exception back to a human.

The same model also extends into household and tabletop settings, where lighting, object positions, materials, and physical interactions are far less predictable.

That requires a bigger "brain" that can perceive, reason, and adapt-not just repeat a narrow set of actions. This is what enables robots to move beyond industrial stations and into real household service.

And it did that for an hour while X Square processed 1,816 parcels at 98%+ accuracy.

Specialization buys you speed.

It still has to prove its broad generalization.

### 引用推文

> X Square Robot：Our livestream has wrapped-and the final result is in: 1,816 randomly selected parcels sorted per hour, with a success rate of over 98%. Since the beginning of ...