Humanoid robots clock 40-second cycles against 15 seconds for a person

Xiaomi said during the 2026 World Internet Conference that its humanoid robot attempted an eight-hour factory shift sorting center-console side covers. It did not confirm that the machine finished it. A published trial at a Chinese car-parts plant puts humanoids at about 40 seconds per cycle against 15 seconds for a human worker.

For factory managers, the gap is now a public number. It is 40 seconds per robot cycle against 15 for a person, at a 98% success rate. That decides whether humanoids take over shifts or stay in pilots. Money is arriving well ahead of the performance data. Agility Robotics expects about $465 million from its SPAC. Nobody has published a full-shift completion rate, an error cost or a payback period.

Xiaomi's eight-hour shift, minus the result

The robot worked a sorting station for center-console side covers. The trial included correcting problems and coordinating with production-line systems. No cycle time for the sorting station was given.

The same disclosure covered a home system built on the Xiaomi-Robotics-1 model. It began trial operations at Xiaomi youth apartments. It completed more than 50 tasks across three rooms in a sequence lasting more than 30 minutes. Xiaomi said the model was pretrained on 100,000 hours of real-operation trajectories. Xiaomi said it ranked first on the RoboCasa 365 and RoboDojo leaderboards in July. Those are training and benchmark claims rather than factory output. An eight-hour shift is the unit that matters to a plant manager. It covers restarts, position errors and the small failures that pile up after the demo ends.

40 seconds against 15

At an automotive parts factory in Zhejiang province, 15 Leju Kuavo robots identified, picked up, transported and placed metal components onto the production line. A human worker needs about 15 seconds for the same task. That makes the robots roughly 2.7 times slower.

The system processed 60,000 components. About 2% of operations were unsatisfactory. That residual error rate is the number a production planner would weigh hardest. A line running at speed still needs someone to catch the failures. Leju ran the robots for about 1,000 hours of real-world operation to collect data and train the AI. The robots adjust their movements when component positions change. That reduces manual intervention. The trial was built to gather data as much as to move parts. Speed is the part that has not moved yet.

Capital runs ahead of the cycle times

More than 84% of votes cast at Churchill Capital Corp XI's special shareholder meeting on 2 November supported the business combination with Agility Robotics. The SPAC is expected to close on 4 November. The combined company is due to trade on Nasdaq from 5 November under the ticker AGLT.

The deal is expected to give Agility about $465 million in gross capital. That is down from the more than $620 million anticipated when the merger was announced in June. The total breaks into about $164 million from Churchill's trust account, about $201 million from a PIPE led by Foxconn, and $100 million in equity Agility received after the announcement. The original transaction valued Agility at a $2.5 billion pre-money equity value. Against that, the company reported $1.8 million in net sales in 2025 in a September S-4 statement. Digit robots were working in 9 customer environments. Agility claims more than $300 million in committed multi-year orders for Digit 5. Its RoboFab plant in Salem, Ore., is designed for annual production capacity of up to 10,000 robots.

Committed orders are not revenue. The multi-year value depends on contractual milestones, product features and specifications, and on customers taking delivery.

Carmakers test, and publish no numbers

Audi is testing Agile Robots' Agile One humanoids at its Ingolstadt Production Lab. The robots pick components from containers and arrange them in the correct order for assembly. The project was announced on 21 October at NVIDIA GTC Berlin. It brings in Dassault Systèmes and NVIDIA. Audi's material supply involves around one million picks daily. That figure describes the existing workflow rather than robot performance. The project has not disclosed quantified performance results or a timeline for wider factory deployment.

Gerd Walker is Audi's board member for production and logistics. He said picking within the sequencing process has remained one of the toughest automation challenges in automotive logistics.

AGCO signed a commercial agreement to pilot Agility's Digit at its Hesston, Kansas plant from 2027. The pilot starts with sorting and stacking parts and placing totes for storage. Agility says the robot will be judged on safety, accuracy, uptime, throughput and return on investment. The pilot has not started. The International Federation of Robotics' World Robotics 2026 report says nearly 7,000 full-size humanoids were sold worldwide in 2025. That is the base the current pilots are building on.

Trimmed forecasts, bigger state bets

Daiwa cut its global humanoid shipment forecasts to 54,000 deliveries in 2026 and 112,000 in 2027. It reaffirmed a buy rating and HK$175 price target on Ubtech Robotics. Media reports cited in that analysis point to Ubtech U1 deliveries falling short of earlier targets. They also point to continuing losses and negative operating cash flow.

Governments are not trimming. South Korea's Korea Institute of Machinery and Materials and the Electronics and Telecommunications Research Institute opened the AI Humanoid Data Factory on 3 November. It runs 12 humanoids across about 1,000 square metres, with plans to expand to 25. Researchers steered the machines by remote control at the opening. They recorded motions that train vision-language-action models. Those systems turn what a robot sees and is told into movement. The Ministry of Planning and Budget earmarked 2.3 trillion won ($1.7 billion) through 2030 for the humanoid industry. That includes buying 1,080 domestic humanoids and raising the local share of core parts to 80% from about 45%. LG Electronics aims to hold about 500,000 cumulative hours of robot training data by next year. That is five times its goal of 100,000 hours by the end of this year.

Data, not hardware, is where much of the state money is going. Whether it closes the cycle-time gap is untested.