Zurich-based mimic robotics has developed a video-action model designed to help industrial robots learn complex manipulation tasks from relatively small amounts of demonstration data. The model, called FLUX-mimic, was developed with Black Forest Labs and combines mimic robotics’ work in robot learning and dexterous manipulation with the architecture of Black Forest Labs’ FLUX 3 video model. The companies are testing the technology in manufacturing environments with Audi.
FLUX-mimic builds on mimic robotics’ earlier mimic-video system, which added robot-action prediction capabilities to a pretrained video-generation model. The new model uses video pretraining to learn how objects move and interact, then applies an action decoder to translate visual predictions into robot movements.
Many robot-learning systems use vision-language-action models trained primarily on static image and text data. Such systems typically require additional robot demonstrations to learn physical dynamics. Mimic robotics said FLUX-mimic can be adapted to some manipulation tasks using about 30 minutes of robot data, depending on the complexity of the task, compared with 30 hours or more under previous approaches.
The company said the reduced data requirement could shorten the time needed to deploy robots for new production tasks. The model is intended for applications involving flexible materials, fine manipulation and product variants that can be difficult to handle with conventionally programmed industrial robots.
Audi is testing FLUX-mimic for production and logistics applications, including tasks involving deformable materials. The carmaker is examining whether learning-based robotic systems can reduce engineering work and support employees by taking over repetitive or physically demanding activities.
Christoph Schneider of the Audi Production Lab said the tests had shown that robots using the model could perform soft-material manipulation tasks that conventional robotic systems had been unable to handle. He said Audi was assessing the technology’s potential to expand flexible automation across its production and logistics operations.
Stephan-Daniel Gravert, co-founder and chief product officer of mimic robotics, said the collaboration would be used to evaluate whether FLUX-mimic could operate reliably on production lines while reducing the engineering work required to introduce new robotic tasks.
Photo: Audi
