Mimic Robotics launches FLUX-mimic with Audi tests

Mimic Robotics has introduced FLUX-mimic, a robot learning model developed with Black Forest Labs. The system is being tested and deployed with Audi.

The Zurich-based robotics company says the model is designed to help robots learn complex manipulation tasks in industrial settings from video, using far less robot demonstration data than many existing approaches.

FLUX-mimic builds on Mimic Robotics' earlier work on Video-Action Models, which adapt video models to predict robot actions as well as images. In this latest version, it combines the company's robot learning and manipulation work with Black Forest Labs' FLUX 3 video model.

Instead of relying on models trained mainly on static image and text datasets, FLUX-mimic starts from a video model that has already learned patterns of motion and behaviour through large-scale video training. It then adds an action decoder that predicts robot actions from visual prediction.

That can sharply reduce the amount of robot-specific training needed for a task. Depending on the difficulty of a manipulation task, the model can be fine-tuned with as little as 30 minutes of robot data, compared with 30 or more hours for earlier methods.

That reduction matters in factories, where industrial robot deployments often require lengthy engineering work, especially for tasks involving flexible materials, soft-body handling or fine manipulation. Those jobs have proved difficult to automate with fixed programming, particularly in production environments with many model variants and frequent changes.

At Audi, the work is focused on those kinds of tasks. The carmaker says conventional robotics has struggled with operations involving flexible parts and precise handling, despite its long history of automation across its manufacturing network.

Robin Rombach, Co-Founder and Chief Executive Officer of Black Forest Labs, linked the project to broader progress in visual AI models.

"Robotics is one of the clearest proofs of visual intelligence. To generate convincing video, a model must learn how the physical world behaves; that same understanding enables acting in it," said Rombach.

"By combining FLUX 3's learned model of the world with Mimic's expertise in robot learning and deployment, FLUX-mimic makes it easier for robots to adapt to new tasks and production environments instead of being engineered for one task at a time," he said.

Factory tests

Audi says it has already moved beyond evaluation and has been testing and deploying the system with Mimic Robotics. The manufacturer is using the work to assess whether learning-based robotics can widen the range of tasks that can be automated across production and logistics.

"In partnership with Mimic, Audi has been testing and deploying FLUX-mimic. We have seen these robots solve complex soft-body manipulation work that would have been simply impossible with conventional robotics. This new technology can have a major impact in assisting our employees, increasing efficiency, and expanding flexible automation across production and logistics operations," said Christoph Schneider of the Audi Production Lab.

"For us, partnering with pioneering companies such as Mimic and Black Forest Labs is essential in pushing the frontier of physical AI and validating these innovations in real-world production environments," Schneider said.

For Audi, the potential appeal is not only automating tasks that have resisted standard robot programming, but doing so without long re-engineering cycles whenever a process changes. Premium automotive production often involves greater variant diversity than mass production, which can make traditional robot cells expensive to rework.

Mimic Robotics aims to address that gap. Founded by ETH Zurich researchers, the company is building both software and dexterous robotic hand hardware for industrial manipulation. Its approach is aimed at more general-purpose automation, rather than robots configured for one tightly defined task.

Black Forest Labs brings a separate set of credentials to the partnership. The AI research company was founded by researchers behind latent diffusion, Stable Diffusion and the FLUX model family, and has attracted more than USD $450 million from investors. It has been valued at USD $3.25 billion.

Deployment focus

Stephan-Daniel Gravert, Co-Founder and Chief Product Officer at Mimic Robotics, said the collaboration with Audi is intended to test whether the model can operate in the kind of demanding production environment where downtime and engineering delays are costly.

"Audi represents the kind of manufacturing partner we built FLUX-mimic for," said Gravert.

"Their production environments demand automation that is flexible enough to handle unstructured tasks, reliable enough for continuous operation, and able to be integrated without months of engineering overhead. Through this partnership with Audi, we want to prove that FLUX-mimic can meet those demands on real production lines," he said.