Abstract
The relationship between my technical capstone paper and my STS research paper is deeply connected, as both look into the integration of artificial intelligence and automation into the traditionally manual environment of American shipyards. My technical project was a proof-of-concept to discover the practicality of this industrial transition. This project focused on the physical and algorithmic development of an autonomous laser robot that is capable of conducting manufacturing tasks. On the other hand, my STS research looks into the social consequences from the integration of what I was essentially building for my technical project. It asks how bringing in this type of technology affects the workers and the social setup of the shipyard. While the technical project views automation as a way to increase precision, safety, and optimization, the research paper critiques this view by examining how technological upgrades can disrupt historical labor hierarchies and worker identities.
My technical capstone project was focused on the creation, modification, and training of a prototype autonomous robot to simulate an automated laser-cutting task in a shipyard environment as independently of human input as possible. Sponsored by Austal USA, the project served as a proof-of-concept for integrating AI into the manual aspects of heavy industrial processes to automate the construction of naval ships. We integrated Hugging Face’s open-source smolVLA (Vision-Language-Action) machine learning model with a LeKiwi robotic arm to automate a simulated manufacturing task. Our end goal was to train the robot to independently identify a red target block from various positions, securely grasp it, and relocate it to a designated green zone. This would allow a simulation of laser detabbing since it includes the act of object recognition, moving with a closed claw to act as a cutting laser, and finishing the cutting by identifying a stop region. The training process was highly iterative; we initially struggled with tensor configurations and hardware limitations that required a dedicated GPU for model evaluation, eventually pivoting our target objects from a 2D circle to a 3D cross, and finally to the red block to improve algorithmic recognition. To adapt the robot for a simulated industrial setting, my team also designed custom printed circuit boards (PCBs) to safely integrate a functional laser module and a warning buzzer, triggered dynamically by the robotic gripper's movements. Through extensive teleoperation, we ultimately developed a 180-episode AI model that achieved a 75% success rate. The project highlighted both the anticipated capabilities and the current computational and environmental limitations of AI models in executing manufacturing tasks.
My STS research paper shifts the focus from machine performance to the societal impacts experienced by the human workforce. Utilizing a socio-professional status framework, the paper argues that the introduction of automated technologies demotes skilled shipyard craftsmen from autonomous professionals to passive machine technicians. Historically, shipyard workers developed their social status and collective trade pride from their craft, relying on specialized knowledge that management did not have the experience for. However, as automation turns technical decision-making into programmed algorithms, this creates a displacement effect that wipes out most machine technicians. Furthermore, designs are also transitioning to a system where components are merely swapped out rather than repaired by experts, meaning workers are stripped of their ability to problem-solve and direct their own labor. Additionally, the research reveals that while management frequently justifies automation as part of improving occupational safety and health, this often masks a strategic effort to break down trade unions and reclaim operational control. Ultimately, the paper concludes that industrial technology is not socially neutral; it acts as a mechanism for redefining who and what holds the power at work.
Working on these two projects concurrently provided a very insightful dual perspective that fundamentally changed my understanding of engineering. Engaging with the frustrating, granular realities of training the smolVLA model, such as overcoming algorithmic bias and organizing datasets to make the robot function as intended, explicitly showed why managers favor machine reliability over human variability. However, experiencing this push for automation firsthand made my STS research much more resonant. I could see exactly how engineering a machine to be perfectly consistent destroys the specialized skills and unpredictability that defines a craftsman's pride and professional identity. In contrast, my STS research also kept me aware of the human cost of the technical efficiency throughout the extensive hours that we spent working on this project. Although we were very cheerful when the LeKiwi robot successfully identified and moved the block without any of our input, my STS framework served as a realistic reminder that in an operational shipyard, this exact success shows that a human worker's experience and skills are no longer necessary. Doing both projects together showed me that engineering has impacts in areas that may not be clear throughout the process of designing. Designing an autonomous system is never just about its software or mathematical technicalities; it takes part in actively reshaping social hierarchies, jobs, and livelihoods.
