The aerospace interior design industry operates under strict regulatory constraints, also requiring rapid development cycles without compromising passenger safety or comfort. Collins Aerospace initiated the project brief to explore how can AI help with the design process. The thesis project explores the main research question “How can AI integration enhance the product development design workflow at Collins Aerospace?” The topic is explored through a contextual research at Collins and three design sprints. Contextual research at Collins revealed an operational gap in the product development workflow: early-stage ergonomic user testing with human participants is rarely conducted due to high financial costs, extended timelines, corporate privacy policies and safety liabilities. As a result, conceptual ergonomic decisions often rely on subjective designer assumptions, self-testing or static industry ergonomics guidelines.

Using the Design Sprints framework (Knapp, Zeratsky, & Kowitz, 2016), it is aimed to discover the topic in depth through three research-design-evaluate cycles. Sprint I explores the AI trends across industries on product development processes alongside the company's contextual research, identifying the lack of human ergonomic user testing as a primary bottleneck. This finding shaped the scope of the project to focusing on improving the user testing aspect of the product development process. Sprint I also included the first ideation of the project. Sprint II explores AI trends and applications in user testing areas both in aviation and in similar industries. Following, the second ideation session is conducted which shaped the primary ideas for the final concept. Later the thesis introduces the final design concept: ErgoSim.

ErgoSim is an AI-augmented, digital ergonomic design review tool that functions as a digital passenger ergonomics twin, integrated directly within the designer's CAD workspace. Operating via a closed-loop 3-tier architecture—connecting a CAD Tier (Rhino/Grasshopper), a Middle Tier (Python and TU Delft DINED database), and an AI Tier (LLM feedback engine), the system extracts spatial dimensions, normalises them against P5Female to P95Male anthropometric boundary envelopes and delivers real-time, context-aware ergonomic reviews.

Later, the ErgoSim tool is tested with five TU Delft IDE students across four aspects: Usability and Navigation, Quality and Trust, Automation Bias, Workflow Integration and Perceived Value. The user evaluation confirmed that ErgoSim successfully transforms ergonomic evaluation from a late-stage review into a real-time, data-grounded co-design process. Participants praised the AI report understandability (4.6 / 5.0) and quality (4.0 / 5.0), highlighting that the AI provided flexible, advisory guidance (4.0 / 5.0) rather than forced automation, thus preserving designer autonomy. While testing identified usability friction points regarding the tool’s canvas navigation and main dashboard settings, ErgoSim demonstrates that embedding generative AI into native CAD environments accelerates iteration loops, introduces complex human factors data and provides objective ergonomics justification for early design decisions. ... Read more

Agent skills provide reusable instruction bundles that guide LLM-based agents through complex workflows. Yet their SKILL.md instructions are written primarily for agent execution rather than non-expert users. Users may know that a skill is available without knowing how to begin, what to provide, or when to adjust its behavior. This thesis asks whether a before-use explanation derived from SKILL.md and reorganized around users' actions can help bridge this gap.

The research followed a two-phase mixed-methods design. In Phase 1, 15 university students used an agent skill and described what they needed before use based on their overall experience. Their difficulties and preferences guided the design of a concise, user-facing explanation derived from SKILL.md. In Phase 2, 24 Master's students took part in a between-subject experiment comparing two pre-task materials: an adapted SKILL.md for the control group and the before-use explanation for the experiment group. Both groups used the same agent skill to complete the same design task. The analysis combined reading behavior, measures of understanding and task planning, interaction logs, and interviews.

The experiment group read the material more completely, reported higher understanding, and demonstrated stronger understanding of available workflow patterns and the pattern used during the task. However, the groups did not differ significantly in task-planning strategies. Qualitative findings suggested that the explanation helped participants recognize input requirements, workflow options, defaults, and checking opportunities, while planning also appeared to reflect prior AI-use routines and task uncertainty. These findings suggest that before-use explanations may help not simply by shortening documentation, but by selecting and reorganizing information that users can act on.

Drawing together the explanation design and empirical findings, this thesis proposes a preliminary translation framework for user-facing agent-skill explanations. The framework highlights three forms of translation: making agent-facing language understandable, connecting procedural knowledge to users’ tasks, and making decision points and opportunities for intervention visible. From a human–AI interaction design perspective, the findings suggest that improving agent-skill use should begin with an information framework that helps users understand the skill and determine how to act, before focusing on interface-level interaction design or visual presentation. ... Read more

Navigation errors (missed, wrong, and risky turns) reflect cognitive failures that remain insufficiently understood in in-vehicle navigation. This thesis examines how level of map guidance detail and cognitive distraction shape these errors through the lens of situation awareness. In a 2×2 within-subjects simulator study (N=40), drivers completed four urban routes under Road-level versus Lane-level Navigation (LLN) guidance, with and without an auditory 2-back task. Multimodal data were collected, including vehicle control, eye movements, secondary task performance, and subjective workload and user experience ratings. Observed navigation errors were mapped to perception, interpretation, or decision-making failures in cognitive processes.
LLN significantly reduced interpretation failures and wrong turns, contributing to a 40% reduction in total errors. It also reallocated attention toward the navigation display, as shown by more frequent and longer glances and broader scanning, without degrading vehicle control. Distraction robustly elevated workload and reduced road monitoring, but session-level error rates remained unchanged. Interaction analyses showed that distraction attenuated LLN’s attention-shift effects, while LLN mitigated some distraction costs in road monitoring; certain control benefits, however, reversed under load. Driver experience moderated outcomes: experienced drivers benefited consistently from LLN, with fewer errors and lower workload, while less experienced drivers reported higher workload and a tendency toward more missed turns.
Together, these findings demonstrate how navigation errors can be systematically mapped to underlying cognitive failures and reveal how level of map guidance detail and distraction influence these processes, providing a foundation for more context-aware navigation support. ... Read more

In today’s world, products are increasingly intertwined with our daily lives, becoming more intricate. To design these interactions effectively, designers need to comprehend both the people involved in the activities and the technical systems. As user-centered design spreads, the focus shifts from viewing design as problemsolving to recognizing it as the social creation of new possibilities. This shift highlights the importance of users and their everyday lives, as well as their interactions with the product in designing.
Videos serve as a powerful tool for learning about users, their behaviors, and interactions. Video Based Design (VBD) involves analyzing videos to interpret, infer, question actions, and understand the user’s thought process. However, this approach can extend the time and analytical rigor needed due to the extensive analysis required to draw valid conclusions from the video content. To streamline this process and enhance researchers’ workflow, we prototyped a tool that can produce mind maps from videos. These mind maps serve as a visual representation of the video’s key concepts in an organized manner. This tool utilizes Large Language Models (LLMs) to process multimodal video inputs and output the generated mind maps.
To assess the impact of mind maps on the video-based design process, an experimental study was conducted involving 28 designers. Participants were asked to watch two videos from different contexts and then engage in activities using mind maps generated by both the LLM and a human designer. Through quantitative and qualitative analysis of the study data, we gained valuable insights, identified strengths and weaknesses, and proposed design enhancements for the next version of the tool. ... Read more

 This study explores integration of Large Language Models (LLMs) and Vision-Language Models (VLMs) into design ideation of industrial design. Conducted at TU Delft, the research involved brainwriting and video-based design (VBD) methodologies. The primary aim was to legitimize and valiate context-injected LLMs and VLMs in supporting designers' search for inspiration through development of experiment framework. The study measured workload, user experience, acceptance of technology, divergent thinking capabilites and attitudes towards AI. It also preliminary analysed the results, focusing on qualitative insights.

Data was collected through surveys and interviews, but although eye-tracking data was also gathered, it was excluded from the analysis. The study found that while AI tools support ideation by generating diverse ideas and handling repetitive tasks, they need improvement for contextually relevant and accurate information. Designers expressed cautious optimism about AI's potential, emphasizing the need for human oversight to retain creativity and ensure context-aware assistance. 

The research highlighted optimistic leaning opinions on AI integration, noting that current AI capabilities are not yet sufficient for design ideation demands. It emphasized the necessity for AI to act as a collaborative partner, preserving the designer's critical role in the creative process.  ... Read more