BMW Group Plant Landshut is expanding its work in physical AI, taking on central software development for AI-supported robotics in component production. The plant's demonstration-based learning workflow uses MANUS data gloves to capture the dexterous manipulation data that trains its humanoid robotic systems.
Plant Landshut is developing a modular robotics ecosystem that combines deterministic programming with Vision-Language-Action models, integrating visual information, language comprehension and robot actions. Getting there depends on reliable training data. The project team captures data from real production environments and test setups and feeds it into robot simulations, where movement patterns and motion planning are modelled virtually before being tested and scaled in stages.
In parallel, humanoid robotic systems acquire new skills through demonstration-based learning. Project team members perform movement patterns and work sequences, which are recorded using camera systems, motion capture suits and MANUS gloves. This data is converted into generalizable behavioral models, referred to as policies, that transfer across different tasks and use cases.
Cameras and motion capture suits handle whole-body posture and spatial context well. Hand and finger articulation is a separate problem, since occlusion and small joint sizes make vision-based systems less reliable for capturing dexterous manipulation. MANUS gloves are the component in Plant Landshut's stack that address that gap, capturing full finger-level degrees of freedom directly from the operator's hand.
This matters directly for Plant Landshut's stated focus areas, which include recognizing and interpreting environments, determining position and status, planning efficient movements, and precisely gripping and moving components. Gripping and moving components is a manipulation task, and manipulation is where finger-level degrees of freedom matter most for policy training. A policy trained on demonstration data with imprecise or under-resolved hand movement is less likely to generalize on any task that depends on fine motor control.
BMW Group Plant Landshut is working with partners including Athenyx Robotics, a startup founded at the Laboratory for Machine Tools and Production Engineering at RWTH Aachen University, and the University of Applied Sciences Landshut, on data generation, simulation and use case development.
