Unitree Robotics G1-D completes 200 factory cycles in an hour; its integrator says people are faster

Text
iHOLYSHITVISUAL / TECHIndustrial robotsiHOLYSHIT / ILLUSTRATION
iHolyShit illustration

Novares reports that a Unitree Robotics G1-D completed 200 autonomous cycles in an hour at its Spanish auto-parts plant. Its integrator says the robot still works more slowly than people. Eleven minutes of task teaching did not cover the whole factory setup.

Automotive supplier Novares released its first results from Seseña, Spain, on November 17. The Unitree Robotics G1-D went into the factory in early November to work alongside production teams. Novares reported 200 autonomous cycles in 60 minutes, with 99.5% success for parts pick-and-place and 95% for box pick-and-place. Those are separate tasks, not one production-line yield. On November 18, Innov8 CEO Stéphane Bohbot told France’s BFM Business that the robot was still slower than a human at repeating the handling work. Improving its pace is the next priority.

The project began with trials at the Lens research center in France before moving to the Spanish plant four months later. BMW offers a reference point from Spartanburg in the United States. In 2024, BMW described a weeks-long trial in which Figure 02 put sheet-metal parts into fixtures. BMW later reported a ten-month pilot there in 2025, with the robot positioning sheet metal for welding. Getting a movement to work on-site and documenting it over a longer operating period are two different milestones.

The team did more than bring a general-purpose machine to the factory

The team did more than bring a general-purpose machine to the factory. Novares defined the work; French integrator Innov8 combined the Unitree hardware with learning software from Bleu Robotics. The system learns from an operator demonstrating a task. Novares says training for the first use case took 11 minutes; Innov8 says the project took three months to prepare. Those figures measure different things: teaching a task and getting a system ready for the plant.

Dividing that hour by the reported cycle count gives about 18 seconds per cycle

Dividing that hour by the reported cycle count gives about 18 seconds per cycle. Bohbot gave no matched human timing, and the average does not establish that the robot could sustain that pace through a shift. If this work feeds another station, a poorly placed part can leave a downstream operator waiting. The questions that matter for staffing are how quickly the system recovers, whether employees must monitor exceptions, and what happens when the material changes.

iHolyshit review I would treat this as an initial result for one operation, not a staffing verdict for a shift. Over the next six to twelve months, watch for records from the same workstation across shifts: time to regain pace after a batch change, how often people intervene, and whether the robot recovers from an error or someone takes over. Only with stable evidence of that kind can the plant compare operating costs.

Share this article

What do you think of this story?

Anonymous feedback, up to 5 reactions per article.

Reader comments

Comment without signing in, or sign in to display your member name.

Discuss the technology and the story. Hate, politics and war, threats, abuse, personal accusations, sexual content, ads, links and personal information are not allowed.

0/500

No comments yet. Be the first to share.

Sources