The post-pandemic period was a critical point for the aviation industry; the shortage of workforce meant difficulties in keeping up with tight aircraft maintenance schedules. To deal with this, KLM Royal Dutch Airlines designed the Back-on-Track program, which emphasized introducing automation as a long-term goal to deal with labour shortages. While there is an attempt to gradually introduce autonomous technologies into the workflow, the opinions and inputs of the employees meant to collaborate with said technologies are not often considered. This gives the impression of a technology push, persuading the employees to collaborate with them. That is where this project comes in; it focuses on technological pull, using the employees’ feedback as input in introducing an autonomous system.

This project is performed at the MRO Lab and aims to aid the aircraft towing operations at KLM’s Hangar 12. KLM uses a remote-controlled towing vehicle called the Mototok to tow an aircraft into and out of the hangar for maintenance. The MRO Lab intends to make this towing process more autonomous, making this project their first attempt to do so.

The project starts with establishing a main research question and sub-research questions, followed by understanding the towing process, specifically its phases, its people, and the interactions occurring between them. To visualize the towing process, a scaled-down, 3D printed version of the setup was created with all towing-relevant infrastructure. An experimental interview arranged with two Mototok drivers helped visualize the towing procedure through the setup, resulting in the development of a process map, containing every step of the towing process, and an interaction map, linking every interaction between operators with the process map.

The first novel aspect of this thesis is a custom framework developed to identify that the safety verification role of the walkers is an ideal starting point to introduce autonomy. This required structuring the flow of information through the Rational Agent Model and breaking the process into functional clusters, followed by the consideration of key inputs from a safety and compliance perspective. Research indicated that the Rational Agent Model matched closely with the Parasuraman, Sheridan, and Wickens framework, which also provides a scale for autonomous system levels. Using its scale as a reference, a Level 2 autonomous system was chosen to replace the safety verification role of the walkers.

Before using the technology in towing, it must be tested. A system-level test was chosen over a product-level test to ensure the autonomy can be evaluated in a context-specific manner. Since the information perceived by the autonomous system would directly impact the decision-making of the drivers, this decision-making was chosen as the test area for the system. However, it was decided to first test the decision-making with the walkers, so that the natural decision-making behaviour could be revealed, along with more specific factors that could be used to test the decision-making of the drivers.

The Critical Decision Method was used to set up the decision-making experiment, which involved an experiment followed by an interview with the driver. The test was based on the driver’s decision-making under certain worst-case scenarios, which they are not trained for, but could occur in towing conditions. A specific situation of the driver receiving multiple inputs was incorporated into the experiment to further understand worst-case decision-making. The scaled-down setup involved simulating the driver and walker’s vision via cameras to maintain a more realistic vision of the towing environment. Results from the scaled-down experiments revealed a decision-making model.

The interview revealed that the driver’s ability to make decisions was heavily reliant on his trust in the walker’s inputs. Thus, the challenge for the perception system test would be to understand if the driver trusts the inputs provided. Guidelines were provided to design this test while addressing any possible limitations.

Finally, the project is concluded and reflected on by answering the main research question and discussing its implications for academia and KLM. The next possible steps to take this project forward are also discussed.