Dr. Häfner, 'AgiProbot' stands for 'Agile production system using mobile, learning robots with multi-sensor technology for uncertain product specifications'. Could you briefly describe the project?

Benjamin Häfner: 'AgiProbot' is a research project funded by the Carl Zeiss Foundation with 3 million euros, in which nine chairs at KIT are jointly researching the possibilities of artificial intelligence and diverse digitalization technologies. This is being done using the example of a production system with so-called remanufacturing - i.e. the disassembly and reassembly - of electric motors.

With what goal?

Benjamin Häfner: The aim of the project is to develop an agile production system that is capable of reacting autonomously to uncertain product specifications. The system to be developed should be able to deal with uncertain component conditions in terms of wear, geometry, degree of contamination, etc. and develop agile solutions to cope with production tasks.

What does the system consist of?

Benjamin Häfner: The agile production system will consist of a series of stations in a matrix factory structure. Each station will consist of a variable number of modular robot systems that work in collaboration with each other or with humans and will be able to adapt to uncertain component properties in a self-learning manner and perform the required disassembly processes autonomously. The intralogistics processes between the stations are also considered by using mobile robotics with autonomous vehicles to enable varying value streams. Various self-adaptive sensor systems are integrated into the plant technology and vehicles. Interaction with humans and the use of existing prior knowledge will also be specifically included in the learning processes mentioned.

As a tangible result of the project, the scientific methods developed will be implemented in an industry-oriented, joint demonstrator factory at KIT, in which the remanufacturing of real electric motors from the automotive industry will be implemented.

What role does artificial intelligence play here?

Benjamin Häfner: In the project, artificial intelligence methods are the key enablers for realizing the necessary learning processes of the agile production system. This concerns different levels of consideration: First of all, the key information on the unknown product features must be captured from measurement data collected using suitable sensor technology. Deep learning methods, for example, are suitable for this. Then the robots have to learn optimal work processes - for example using various reinforcement learning approaches. Assisted learning through human observation is also being intensively investigated. Finally, the material flows in the matrix factory structure of the production system should also be learned autonomously, which is possible using so-called Q-learning, for example.

What are the challenges?

Benjamin Häfner: The particular challenge of the project is to design the aforementioned approaches across domains and implement them together in a realistic factory structure in order to develop a dynamically learning overall production system. This leads to significantly greater complexity than if the various subsystems are only considered independently of each other.

What does the project roadmap look like?

Benjamin Häfner: We are organizing the project based on SCRUM principles, which are already widely used in the field of software development. This is characterized in particular by the fact that we are developing iteratively as part of the project in order to make the complexity as manageable as possible. Over the course of the five-year project, we have set ourselves the goal of implementing a simple but functional version of the agile production system after the first year.