Humanoid Robotics at Landshut
BMW expands humanoid robotics software at Landshut plant
BMW is establishing its Landshut plant as a centre for humanoid robotics software development, advancing AI models, simulation and robot training to support future manufacturing applications alongside pilot projects in Leipzig and Spartanburg.
BMW is building up expertise for humanoid robotics at the Landshut plant. In future, the site will take on central development tasks for the software of the systems. These include the generation and preparation of training data, virtual simulations, motion planning and the training of the robots. In this way, Landshut complements existing pilot projects at the BMW plants in Leipzig and Spartanburg .
While concrete applications in the production environment are being tested there, the components plant is primarily intended to create technological foundations for a later broader deployment. “At the BMW Group plant in Landshut, we combine software expertise with the industrial practice of our component production. This allows us to test new technologies at an early stage and to assess their benefits for production on a sound basis,” says Wolfgang Blümlhuber, Head of Technology Vehicle Dynamics and In-house Component Production at BMW.
The focus is on activities that can only be mapped with difficulty or at high cost using classical automation. These include tasks with changing sequences, varying component positions or fine motor requirements.
The robot systems are to perceive their environment with the help of sensor and image data, assess situations and derive suitable movements from this. BMW wants not only to investigate the technical feasibility, but also to assess in which processes humanoid robots can offer an economic benefit.
AI models and simulations
Technologically, BMW is relying on an open, modular software architecture. It is intended to combine firmly programmed sequences with AI models. An important role is played here by vision-language-action models, which translate visual information and task descriptions into concrete robot movements.
Before applications make their way into ongoing production, they are first trained in simulations and test environments. Cameras, motion-capture suits and data gloves, among other things, are used for data collection. “For us, it is crucial to develop the technology very close to real requirements. We are not only examining what is technically possible, but also where humanoid robotics in component manufacturing can actually bring a robust benefit,” says Christoph Jagoda, project manager Physical AI at BMW Group Plant Landshut.
In development, BMW is working together, among others, with Athenyx Robotics and Landshut University of Applied Sciences. The aim is a so-called intelligence stack that combines perception, learning, simulation and decision-making. In the long term, BMW wants to transfer learned capabilities to different robot models and production applications. What remains decisive for later scaling is how reliably the systems operate under changing production conditions.
