Humanoid Robots Are Now Sorting Packages at China's Postal Hubs. A Human Still Beat One by 192 Parcels

RobotEra M7-style humanoid upper-body robot sorting parcels on a warehouse conveyor belt

1,200 parcels per hour. That is the number China Post Group is putting forward for its newly deployed humanoid robots at the Guangzhou Jianggao logistics hub. In the same week, a US startup ran a 10-hour head-to-head between its humanoid robot and a human intern. The human won. By 192 packages. These two data points, taken together, tell you more about where humanoid package sorting actually stands than any individual press release does.

What China Post Actually Deployed in Guangzhou

China Post Group Co. Ltd., the country's state-owned postal carrier, has installed humanoid robots at its Jianggao facility in Guangzhou, one of the busiest logistics hubs in southern China. Photos published by People's Daily, the state media organ, show robotic sorters working alongside robotic arms and unmanned forklifts on the warehouse floor.

The facility is not small. It processes an average of 6.5 million parcels per day, with peak volumes exceeding 10 million. The humanoid robots are one layer inside a much larger automated system, not the whole system. That context matters when evaluating the deployment.

How many robots were installed? The announcement did not say. "An unspecified number" is the honest answer.

The RobotEra M7: What It Is and What It Is Not

The robot in question is the Xingdong M7, made by Beijing-based startup RobotEra. According to Chinese social media reporting cited by Futurism, the M7 is what you might call a partial humanoid. The base unit is not bipedal. It is a torso mounted on a fixed stand, not a walking robot. Calling it "humanoid" is technically defensible because of its upper-body design (arms with 7 degrees of freedom, hands with 12), but it will not be walking out of the warehouse anytime soon.

What the M7 does have is a 360-degree field of view integrated with 3D LiDAR, which it uses to locate and handle packages arriving on the conveyor belt in front of it. The robots are stationed at fixed points along the belt. The conveyor brings the packages to them.

Note: The M7 can be upgraded to include legs, per RobotEra's product page. The units currently deployed at China Post are the fixed-base version. Walking capability is not part of this deployment.

RobotEra is a 2023-era startup that has raised at least $400 million to date, with SF Express, one of China's largest logistics companies, leading a major funding round. The company says its robots are now operating at more than 10 logistics centers across China. The M7's companion model, the fully bipedal L7, is also being piloted at some facilities, though the Guangzhou deployment uses the fixed-base M7. For a broader look at how different humanoid platforms are approaching the same deployment challenges, see 3 humanoid robot research platforms compared.

The 1,200 Parcels Per Hour Claim: Read It Carefully

People's Daily states that the robots together are capable of processing up to 1,200 parcels per hour. A few things to parse here. First, "together": this is a combined throughput figure for all the deployed units, not per robot. Since the number of robots is undisclosed, it is impossible to derive a per-unit efficiency number from this claim. Second, "capable of" and "up to" are ceiling figures, not average operating rates.

This does not mean the number is misleading. A combined throughput figure at a state-run facility is a reasonable way to report operational capacity. But it is worth understanding what the claim does and does not tell you before treating it as a benchmark.

No error rate, accuracy figure, or comparison against the facility's human sorters was included in the announcement.

Meanwhile in the US: A Human Beat a Robot by 192 Packages

Around the same time as the China Post announcement, US robotics firm Figure AI ran a 10-hour livestreamed contest between its F.03 humanoid robot and an intern named Aime Gerard. The task: sort as many packages as possible from a conveyor belt. The intern got paid breaks and lunch, as required by California labor law. The robot ran autonomously with no teleoperation.

Final result: Gerard sorted 12,924 packages at an average of 2.79 seconds each. The robot sorted 12,732 packages at 2.83 seconds each. The gap was 192 parcels and 0.04 seconds per package. Figure AI CEO Brett Adcock posted afterward: "This is the last time a human will ever win."

That is a remarkable result. Not because the robot lost, but because the gap is now this small. Earlier demonstrations of humanoid package sorting were not competitive with human throughput at all. A gap of 0.04 seconds per parcel, across a 10-hour sustained run with full autonomy, is a different category of performance than where this technology was 18 months ago.

Caveat: The Figure AI test used a single robot on a controlled task in a company-run environment. Independent verification of the throughput figures, error rates, and recovery events was not provided. Deployment at a customer facility under real operational conditions is a different challenge.

Why the People's Daily Spotlight Matters More Than the Specs

China's robotics push is not happening in a vacuum. The country broke the record for industrial robots in operation in September 2025, and produces roughly 90 percent of the world's humanoid robots. Factories, kiosks, and logistics hubs across China are running pilots with humanoid systems from multiple companies. The software infrastructure driving these deployments is equally significant. NVIDIA's physical AI stack, covered here in detail, is one layer several Chinese manufacturers are building on: NVIDIA's Physical AI Stack: Cosmos 3, Vera, and the Robot Operating Layer.

The fact that People's Daily, a state media outlet, published photographs of the Guangzhou deployment is itself a signal. Chinese state media does not run photo-ops for technology it is skeptical of. This is official endorsement of the country's robotics industry, placed inside a highly visible state-owned infrastructure operator.

Whether the M7 at China Post is currently more efficient than a human sorter is a secondary question. The primary message from the deployment is institutional confidence, not operational superiority.

My Take

The China Post deployment and the Figure AI intern test are being reported as separate stories. They are the same story. Humanoid package sorting has moved from demos to actual operations. The gap between robot and human throughput is now measurable in fractions of a second, not orders of magnitude. And the world's largest logistics operations are betting real infrastructure on these machines.

That said, "capable of 1,200 parcels per hour combined" from an undisclosed number of robots, measured by state media, is not a rigorous benchmark. It is a headline. The Figure AI number is more useful precisely because it was a controlled, documented contest with actual per-unit data, and even that came without independent verification.

The gap is closing fast. Whether it closes all the way, and at what cost per robot compared to a human worker, is the question nobody in these announcements is actually answering yet.

Key Takeaways
  • China Post has deployed RobotEra M7 robots at its Guangzhou Jianggao facility: fixed-base torso units, not walking humanoids.
  • The combined throughput claim is 1,200 parcels per hour across an unspecified number of units. No per-robot or accuracy data was published.
  • In a head-to-head test, Figure AI's F.03 robot sorted 12,732 packages in 10 hours; a human sorted 12,924. The gap: 0.04 seconds per parcel.
  • People's Daily coverage of the China Post deployment is a state-level endorsement signal, not just a product announcement.
  • The facility processes 6.5 million parcels daily on average. The humanoid robots are one part of a larger automated system.

FAQ

What robot is China Post using to sort packages?

China Post's Guangzhou facility is using the RobotEra Xingdong M7, a fixed-base upper-body robot with 7-degree-of-freedom arms, 12-degree-of-freedom hands, and a 360-degree field of view with 3D LiDAR. It is not a walking humanoid; it operates from a stationary stand positioned along a conveyor belt.

Can humanoid robots sort packages as fast as humans?

In a controlled 10-hour test by Figure AI, the gap was 0.04 seconds per package. The robot averaged 2.83 seconds, the human averaged 2.79 seconds. The human won by 192 packages out of roughly 12,900 sorted. That is near-parity on a single controlled task, though real-world deployment conditions involve more variability than a structured contest.

How many parcels does China Post's Guangzhou hub process per day?

According to Chinese state media, the Jianggao logistics facility in Guangzhou handles an average of 6.5 million mail items per day, with peak volumes exceeding 10 million. The humanoid robots work alongside robotic arms and unmanned forklifts as part of a broader automated system.

Who makes RobotEra robots and where are they deployed?

RobotEra is a Beijing-based robotics startup founded in 2023. The company has raised at least $400 million, with SF Express, a major Chinese logistics firm, as a lead investor. RobotEra says its humanoid robots are now operating in more than 10 logistics centers across China, with China Post and SF Express as key partners.

Are humanoid robots replacing human workers in Chinese warehouses?

Current deployments, including the China Post facility, show humanoid robots working alongside existing automated systems: robotic arms, forklifts, rather than replacing all human workers outright. How many human roles are affected at the Guangzhou hub specifically was not disclosed in the announcement.

Conclusion

The China Post deployment and the Figure AI intern contest happened in the same week. That timing is not a coincidence. It reflects where the whole field is right now. Humanoid robots are moving off lab floors and into real logistics operations. The throughput gap with human workers has collapsed to fractions of a second on controlled tasks. Whether that translates to consistent, cost-effective performance at the scale of millions of packages per day is still an open question, and the one nobody is rushing to answer publicly. For more on the hardware platforms being built for exactly these environments, see NVIDIA's humanoid robot platform built around Unitree.

Post a Comment

0 Comments