(Yicai) June 29 -- AgiBot has delivered its 15,000th humanoid robot less than three months after the 10,000th. The milestone arrived as the Chinese startup wrapped up a successful six-day live-streamed factory trial, during which its Genie G2 robots inspected tablet computers on a commercial production line.

“Shipping out the 15,000th robot marks a new milestone in mass production and shows China's humanoid robotics industry is moving into large-scale deployment in real-world applications," Yao Maoqing, a partner at AgiBot and president of the Shanghai-based firm’s embodied intelligence business, said yesterday.

Established in February 2023, AgiBot sold its 10,000th humanoid robot in March. It delivered 5,168 last year for a 39 percent global market share, ranking first worldwide in both categories, according to market research firm Omdia.

“Most of my day is spent coordinating deliveries, supply assurance, quality control, and engineering changes,” said Yao, highlighting the operational strain on AgiBot’s rapid scale-up. He cited one example in which a quality engineer was stationed at a supplier producing a key Genie G2 component for more than three months, helping boost production capacity fivefold while raising the first-pass yield to over 95 percent from less than 60 percent.

AgiBot has also built a standardized supply chain ecosystem across the Yangtze River Delta, with more than 70 percent of components now sourced from the region, the company said.

Stress Testing

The livestream was intended as a public stress test of the Genie G2 operating on an active production line belonging to Longcheer Technology in Nanchang, Jiangxi province. Shanghai-based Longcheer is an original design manufacturer that counts Xiaomi, Samsung Electronics, and Lenovo among its clients.

The Genie G2 robots worked from 8 a.m. to 7 p.m., in line with the factory's production schedule, picking tablets from conveyor belts and inserting them into precision testing equipment at final inspection stations to sort qualified and defective products before shipment. After inspecting more than 17,000 tablets, they had a success rate of 99.99 percent.

The robots had been operating on a secondary production line of Longcheer’s since last December before being integrated into the main production line in March.

The livestream was not without friction. Early in the session, one robot suffered a communication failure with test equipment, interrupting the inspection cycle and causing two tablets to stack together, requiring manual intervention. The incident briefly reduced the success rate to around 99.96 percent. The robot immediately reported the fault, switched to another station, and resumed operation about two minutes later without substantially affecting overall production.

"I hope every problem shows up during the livestream, because every issue we uncover brings us one step closer to large-scale deployment," said Li Long, general manager of the robotics business department at Longcheer.

The livestream was shown across multiple domestic and overseas platforms. Some overseas viewers compared it with Figure AI's earlier 200-hour live warehouse-sorting demonstration, but noted that deployment on a third-party’s active production line presents far higher variability and operational complexity.

Agibot said that over the period from initial deployment on Longcheer’s secondary line through full integration in March and into the livestream phase, a joint engineering team identified and resolved about 60 issues. These included sensor misreads that led to robot collisions and communication protocol mismatches that resulted in lost commands.

Ensuring Client Payoff

Following on from mass production, ensuring stable robot performance in customer environments has become AgiBot’s next major challenge. “Robots need to match the pace of human workers while maintaining a very high level of reliability,” Li said. “That's the key problem we still need to solve.”

Their artificial intelligence systems have not yet reached the point where they can learn a new task simply by seeing it once, with every new application still requiring pre-training and post-training by algorithm engineers, Li noted.

Robots also operate as part of a broader production system involving dozens of workers, meaning a single malfunction can disrupt an entire production line, Li pointed out.

AgiBot's goal is to achieve a "deployment-ready" state, where robots can be put to work immediately rather than functioning as development platforms, Yao said. "Like buying a car, you should be able to start using it right away.”

To bridge the gap between manufacturing and deployment, AgiBot is building a partner ecosystem, including training programs that allow external developers to maintain and debug systems independently, alongside continued refinement of its software toolchain.

It has also launched Genie Studio Agent, a development platform enabling partners to build customized applications and integrate robots into customer workflows while lowering deployment barriers.

Bulk-Buying Shift

Orders from manufacturers are shifting from pilot projects to bulk procurement, AgiBot noted, adding that it expects to deliver more than 1,000 robots by year-end across semiconductor production, auto components, logistics and warehousing, and commercial services.

In March, AgiBot's humanoid robots were deployed at the battery production line for the Buick Electra E7 at SAIC-GM's Ultium plant, performing high-precision operations with positioning accuracy of plus or minus 0.1 millimeters. This was the first android used by Chinese carmaker SAIC Motor on a mass-production line and among the first in China's auto industry.

At the BEYOND Expo held in May, AgiBot said it was continuing to expand overseas. Overseas income accounted for less than 10 percent of revenue last year, but doubled to nearly 20 percent in the first half of this year. The company is targeting North America, Europe, Japan, and South Korea as its main foreign markets.

Global demand for humanoid robots could reach about 86 million units by 2050, creating a market worth between USD1.4 trillion and USD1.7 trillion, according to Swiss bank UBS.

But the industry's "electric vehicle moment” -- where annual humanoid robot sales surge from millions to tens of millions within a short period -- remains unlikely before 2030 due to immature AI capabilities, insufficient training data, and the absence of mature regulatory frameworks, said UBS analyst Phyllis Wang.

Editor: Martin Kadiev