The aviation industry is increasingly relying on automation and manufacturing robots to address the increase in production rates as well as the shortage of skilled workers. In contrast to the automotive industry, production volumes in aircraft manufacturing are relatively low, so establishing rigid production lines would be inefficient. The AIARA research project, therefore, investigates the possibility of using manufacturing robots with minimal effort in different areas. AI-based approaches should enable them to independently acquire new tasks.

In terms of endurance and precision, manufacturing robots far surpass their human colleagues. Day by day, they perform their tasks step by step without tiring. However, they have been dependent on strictly programmed routines so far: grabbing the screw, inserting the screw, tightening the screw. A change in the component or the screw position required manual reprogramming.

AI-based approaches like ‘deep reinforcement learning’ now open up new possibilities. ZAL development engineer Stephan Rediske explains the new method: “Instead of following rigid instructions, the robot can in the future ‘understand’ its task and independently find the best solution. Through continuous experimentation, the robot learns which movements bring it closer to its goal. Successful steps are rewarded virtually by us.”

Use Case: 24-Hour Operation in the 3D Printing Lab

As part of the AIARA project, a practical use case from the field of 3D printing was studied. 3D printers play a significant role in aviation research, especially in prototype manufacturing, and could also contribute to a more personalized design of aircraft cabins in the future (see project LiBio)

Currently, so-called desktop 3D printers can produce parts around the clock but require a human operator to remove the finished workpiece from the printing bed after the printing process. This is often not possible, particularly at night. To ensure continuous 24-hour operation, a robot could automatically move between the printers, open the printer door, and remove the finished workpiece. An autonomous robot offers the advantage of keeping production flexible and adaptable at any time. Even with a shift in the position of a printer, the robot can use sensors to recognize the new position of the door knob and perform its task.

Virtual Classroom

However, the robot learns its new task not in the real world but in a virtual training environment (Gazebo from Open Robotics). This not only guarantees safety and full control of all influences but also enables the simultaneous simulation of many different training variants in parallel. Its virtual classroom is a digital twin of the real AIARA demonstrator: Through a camera, it sees the printer mockup and tries to guide its gripping arm to the door knob. Its “teacher” is a so-called reward function. If it brings the tips of its gripping arm close to the knob, this path is considered successful. If it is too far away or collides with the printer cell, it has failed. The inertia and weight of the materials are also calculated to ensure the results can be transferred to reality. In the demonstrator setup, the virtual and real test environments were successfully synchronized.

Future Robots Controlled by Voice Command?

That more autonomous robots are given high priority is also demonstrated by other spectacular publications: The manufacturer Figure presented a humanoid robot in spring 2024 that could understand and solve complex tasks using speech recognition on the spot. A collaboration with BMW is forthcoming, where it will demonstrate its capabilities. How autonomous robots will fare in a real production environment remains to be seen. However, projects like AIARA show that manufacturing scenarios can be very well mapped in a simulation environment without requiring high-performance hardware. There are still many open questions regarding reliability and safety, but undoubtedly, robotics and AI will play a key role in the aviation industry for tomorrow’s manufacturing.

Are you interested in manufacturing simulations and AI or have questions about the AIARA project? Then share your opinion with us and give us a call:

Project partner:

• Broetje-Automation GmbH (Rastede, Deutschland)
• DLR ZLP, Augsburg
• Fraunhofer-Institut für Produktionstechnologie IPT, Aachen
• Kinova
• University of British Columbia