What happened
Agile Robots’ announcement, published on October 27, 2026, says the company and NVIDIA launched Agile ONE SN at NVIDIA GTC Berlin on October 21. The company describes it as a humanoid reference platform for research and development, built on the NVIDIA Isaac GR00T development platform. The robot is 1.4 metres tall and weighs 43 kilograms. It has two WUJI tactile five-finger hands and 71 degrees of freedom, runs on NVIDIA Jetson Thor and ships with the GR00T 1.7 open vision-language-action model. Agile Robots says the platform is intended to let research labs develop skills that can transfer to the full-size industrial Agile ONE. Availability is planned for early 2027; the announcement does not state a price.
The release also describes a joint project involving Agile Robots, AUDI AG, Dassault Systèmes and NVIDIA at Audi’s plant in Ingolstadt. In the Audi Production Lab, mobile humanoids are to pick parts from containers and sequence them for final assembly. Agile Robots says this task is performed around one million times a day. The workcell is first tested in a DELMIA virtual twin accelerated by NVIDIA Omniverse libraries. The announcement presents this as a project, not a report of a completed factory deployment or measured production results.
Agile Robots says it has released new-generation Diana 7 and Diana 3 arms, both supporting the Franka Control Interface (FCI). The company states that both share ±0.05 mm repeatability and 0.5 N relative force-control accuracy, and that Franka Research 3 code can move directly to Diana through FCI. Diana 3 retains seven degrees of freedom, joint torque sensing and compliant control, and is positioned for precision manipulation, lab automation and sample handling.
Diana 7 extends the Franka Research 3 to higher payloads and reach, according to the announcement. It offers a 7 kg payload and 923 mm reach, and weighs 2 kg less than before. Agile Robots says it showed Diana 7 at the 26th China International Industry Fair in Shanghai, held October 12–16. The release does not give pricing or availability details for either arm. It describes Agile Robots as founded in Munich in 2018 by robotics researchers from the German Aerospace Center and says the company employs more than 3,500 people worldwide.
What the archive adds
New Runtime’s earlier coverage on training robots with human videos and simulated practice (October 22, 2026) described Agile Robots’ approach to addressing robotics data scarcity. Its coverage of a planned 1,000–2,000 small humanoids with NVIDIA (October 30, 2026) covered a production target. This is a new product and factory-project announcement, adding specifications and a named use case to a known company story.
Trend signal
This supports generative media becoming application infrastructure only weakly: the announcement connects a model and simulation tools to a physical application, but it does not establish a repeatable infrastructure layer or report measured results. This does not yet fit a pattern New Runtime tracks more directly.
Why now
The release places the humanoid launch at NVIDIA GTC Berlin on October 21, shortly after the announcement of a separate Audi Production Lab project. It also says NVIDIA announced plans to invest in Agile Robots two days before the launch, at the Invest in Germany Summit in Berlin. The timing brings the platform, industrial application and investment plan into the same week of announcements; the release does not specify the investment amount or terms.
Connections to other events
New Runtime’s earlier coverage of the planned Agile Robots production target connects the new research platform to the company’s stated interest in scaling humanoid manufacturing. The Audi project adds a specific logistics task, while the earlier note on training data described a separate part of the robotics development challenge. The release does not say whether the Audi workcell will use skills trained through the human-video approach or the Agile ONE SN platform.
Why it matters
For teams building physical AI systems, the announcement makes a useful distinction between a research platform and an industrial deployment claim. Agile ONE SN is presented as hardware for developing transferable skills; the Audi workcell is a project whose first step is testing in a virtual twin. That pairing suggests a path for evaluating a task before using robots on a production line, but the announcement provides no performance data showing that simulation predicts reliable operation on the factory floor.
The arm details also matter to robotics teams maintaining software across hardware generations. Agile Robots says FCI support lets Franka Research 3 code move directly to Diana 3 and Diana 7. If that compatibility works as stated, it could reduce porting work for existing users. Teams should treat this as a vendor claim to validate against their own controllers, payloads and precision requirements, not as proof that every application transfers unchanged.
For now, the practical value is strongest for research labs evaluating a humanoid platform or manufacturers investigating a tightly scoped logistics task. The release gives dimensions, degrees of freedom, payload and accuracy figures, but no price, uptime, cycle-time, safety or comparative benchmark data. Those are the measures buyers would need before deciding whether the hardware is ready for their operations.
What happens next
- Agile Robots will begin offering Agile ONE SN to research users. (3 months, 60%). Watch for: a confirmed availability date and ordering details.
- The Audi project will report a completed virtual-twin test before making a production claim. (6 months, 45%). Watch for: published task-completion, reliability or cycle-time results from the Ingolstadt workcell.
Limits and uncertainty
The announcement is a company release and its product specifications and project descriptions have not been independently checked here. Agile Robots does not disclose Agile ONE SN pricing, detailed ordering terms or the exact meaning of planned early-2027 availability. It gives no measured task success rate, cycle time, uptime, safety performance or comparison with human workers for the Audi project. The release says the workcell is first tested in a virtual twin but does not say when a physical factory trial will begin, how long it will run or what criteria will determine success. It also does not quantify the planned NVIDIA investment.
Skill to build now
If your team is considering humanoids for logistics, turn the target operation into a testable workcell specification this week: define the part variants, container locations, required sequence, allowed cycle time and failure conditions. Use that specification to compare a simulated workflow with a physical trial when one becomes available. The Audi announcement names picking and sequencing but gives no success criteria, so teams should establish their own before evaluating claims.
Reader takeaway
Agile Robots has announced a research humanoid with concrete hardware specifications and a factory project aimed at a repetitive Audi logistics task. The important distinction is that a named project and simulation-first workflow are not yet proof of production readiness. For builders, the next meaningful evidence will be transferable skills, physical-trial results and operational measures such as reliability and cycle time.
Open question
Will skills developed on Agile ONE SN transfer to the full-size industrial robot under real factory conditions, and how will Agile Robots measure that transfer?