Schaeffler and German robotics software developer Wandelbots have agreed to develop a physical AI platform based on Wandelbots NOVA for use across Schaeffler’s global manufacturing network. The partners intend to connect robot applications, simulation models and artificial intelligence through common software, reducing the effort required to develop and transfer automation between factories.
The technology partnership was announced on 22 October during NVIDIA GTC in Berlin. Schaeffler aims to make the resulting platform available to its plants worldwide and is targeting a reduction of up to 40% in automation costs over the equipment lifecycle. That percentage is an objective for the partnership, not a saving already demonstrated across Schaeffler’s production estate.
NOVA provides a software layer through which supported industrial robots can interact with applications built using common interfaces. Schaeffler intends to combine it with its own production engineering knowledge, internally developed functions and third-party software. NVIDIA Omniverse and Isaac Sim will provide virtual environments for testing equipment arrangements and robot behaviour before deployment.
The programme extends previous work rather than introducing the companies to one another for the first time. In an earlier robotic packaging cell project, Schaeffler used NOVA and NVIDIA simulation to examine movements, cycle times and inspection operations in a digital representation of the production cell. Wandelbots reports that this approach reduced development time by approximately two weeks per application. It does not provide a network-wide comparison of commissioning costs or specify a standard baseline for all robotic tasks.
Industrial robot software is often closely associated with a particular controller, machine type and application. Even where two robots carry out similar operations, differences in programming interfaces, available motion functions and safety systems may require engineers to adapt significant parts of a previously developed solution. Those tasks become more expensive when a manufacturer operates different generations of equipment across numerous sites.
The NOVA approach separates higher-level application logic from at least some hardware-specific communication and execution functions. Software developed for a supported configuration can then be tested and adapted for another installation without rebuilding the entire application from the beginning. Compatibility remains specific to controller models, hardware capabilities and integration arrangements, rather than extending automatically to every robot.
A production cell built around one type of manipulator may rely on a particular reach, payload, gripper design and working envelope. Moving the same process to a different system can change the available positions and the route between them. Shared software may reduce programming effort, but engineers must still verify the geometry, movements and operating limits of the physical equipment.
Virtual simulation helps address some of those differences before the production line is changed. A digital cell can represent robot geometry, tooling and the surrounding machinery, allowing engineers to examine reachability, potential collisions and the order in which operations occur. The accuracy of those checks depends on the quality of the model and its correspondence with the intended physical installation.
The partners intend to use the same NOVA software interfaces for virtual and physical systems. This can allow application logic developed during simulation to be carried into commissioning with fewer changes. Practical installation still requires calibration, safety validation and checks under operating conditions, particularly where sensors, material variation and mechanical tolerances affect the task.
The proposed Schaeffler platform also includes a centrally managed library of reusable robotic skills. A skill may represent a defined sequence, such as positioning a part, moving an item or completing an handling operation associated with an inspection. Once developed and tested, software and associated process knowledge could be shared with other plants rather than retained within an isolated engineering project.
Reuse does not remove the need to qualify a skill at its new destination. An item may arrive in a different orientation, a gripper may have different contact characteristics, or a local machine may impose another sequence of interlocks. The library can supply a starting point and a documented application history, but each deployment must meet its own operational requirements.
Physical AI adds the potential for software models to interpret observations and suggest or generate actions where deterministic programming is less flexible. Connecting such systems to factory machinery requires boundaries between model outputs and permitted physical movements. An AI-generated action must be compatible with the robot controller and the defined operating conditions before the equipment executes it.
Wandelbots describes NOVA as an execution and development environment with tools for testing and controlling supported robots. The companies intend to extend its use to industrial robots, mobile systems and potentially humanoid applications. Schaeffler has not specified which equipment categories are already supported within its factories or provided a timetable for adoption across the manufacturing network.
Schaeffler’s 20 October opening of a humanoid robotics factory in Taicang, China, illustrates a separate investment in robot components and production research. The NOVA partnership concerns common application software and deployment methods; it should not be presented as a second announcement of that factory or as evidence that all Taicang systems already operate on the new platform.
The partners have not disclosed the investment value, the first factories scheduled for the expanded platform or a date for full implementation. Its performance will be demonstrated through repeated use of applications, reduced engineering and commissioning work, and reliable transfer of robot functions between supported production systems. Those measures, rather than the unverified 40% target, will determine the realised manufacturing benefit.




