BMW is developing its own software platform for future humanoid factory workers. At its Landshut component plant, the carmaker is developing a modular software stack that connects production data, artificial intelligence, and robots from different suppliers.
Rather than building a BMW-branded rival to Tesla’s Optimus or XPeng’s IRON, it wants to control how humanoid robots are trained, integrated, and deployed across its factories.
Landshut will work on perception, motion planning, simulation, and Vision-Language-Action models, which translate images and instructions into physical actions. Workers can demonstrate a task using cameras, motion-capture suits, or data gloves, after which a robot learns a reusable skill.
Athenyx Robotics
BMW is working with Athenyx Robotics, a spin-off from RWTH Aachen University and Landshut University of Applied Sciences. The aim is an open architecture compatible with several robot brands. Skills learned on one machine could eventually be transferred to another application without starting from scratch.
That is strategically important. The most valuable asset may not be the robot’s body, but the production data teaching it how BMW components should be recognized, handled, and assembled. By controlling that data and the integration software, BMW can reduce its dependence on a single supplier and protect its manufacturing knowledge.
Landshut, BMW’s largest component plant, makes a wide range of parts for BMW, MINI, Rolls-Royce, and BMW Motorrad. Its varied production offers suitable tasks for robots.
Working with robots
BMW already has experience with humanoids. Figure 02 worked for ten months at the Spartanburg plant in the US, handling sheet-metal parts for BMW X3 production.
BMW says it moved more than 90,000 components during around 1,250 operating hours and contributed to more than 30,000 cars. Figure 03 is now being introduced for a more complex logistics-sequencing task.
BMW is also testing Hexagon’s AEON robot at Leipzig. AEON combines a humanoid upper body with a wheeled base and is being evaluated for logistics, component production, and high-voltage battery assembly. Using two suppliers underscores BMW’s intention not to lock itself into a single platform.
Humanoid robots interest carmakers because factories are designed around the size, reach, and movement of people. Unlike conventional industrial robots, which are usually fenced off and programmed for a single fixed operation, humanoids can use existing aisles, workstations, and tools. They may automate low-volume or variable tasks that do not warrant a dedicated installation.
BMW says the machines should support employees by taking over repetitive, physically demanding, or dangerous work. Reliable robots could also address labor shortages, extend operating hours, and make mixed production of combustion cars, plug-in hybrids, and EVs more flexible.
Other carmakers invest in robots too
BMW is not alone. Mercedes-Benz is testing Apptronik’s Apollo robot for intralogistics and repetitive tasks. Hyundai Motor Group owns Boston Dynamics, whose production-ready Atlas is being prepared for automotive material handling.
Tesla and XPeng pursue a more vertically integrated strategy. Tesla develops Optimus hardware, actuators, AI software, and its own production. It says Gen 3 should enter production before the end of 2026, although its eventual target of one million units annually remains highly ambitious.
XPeng also sees humanoids as an extension of automotive AI. Its next-generation IRON uses the company’s Physical AI and Vision-Language-Action technology, and XPeng aims for large-scale production by the end of 2026. It wants robots to become a business alongside cars and robotaxis.
Cautious route
BMW is taking a more cautious route. It is not announcing robot sales or its own humanoid hardware. Its immediate goal is to become an expert user and systems integrator, able to select the best machines, teach them BMW-specific tasks, and transfer successful applications across its global production network.
BMW’s deployments involve narrowly defined operations, not autonomous general-purpose workers, and it has not disclosed the number of robots or a timetable for widespread use.
Landshut nevertheless shows where the real race may be heading. It is not only about building a humanoid body, but about owning the data and intelligence that make it useful on a factory floor.
