Abstract
This portfolio brings together two major components of my undergraduate work: a technical capstone project focused on radar-based SLAM for ego-motion estimation, and an STS research paper analyzing the ethical and societal implications of machine perception systems. While these projects approach the same general domain from different angles, they are closely connected through a shared focus on how machines perceive, interpret, and act within the real world.
My technical project centers on building and evaluating a radar-based SLAM pipeline for estimating the motion of a moving platform. The motivation for this work comes from the growing importance of autonomous systems, especially in areas like self-driving vehicles and robotics, where reliable perception is critical. Unlike camera or LiDAR-based systems, radar offers advantages in challenging conditions such as rain, fog, or low visibility, making it a valuable but more complex sensing modality.
In this project, I worked with radar datasets to reconstruct motion by processing raw sensor outputs into meaningful spatial representations. The pipeline involves several key steps, including transforming radar signals into usable point clouds, grouping data into time-based frames, and matching points between consecutive frames. I implemented algorithms such as the Kabsch algorithm to estimate rotation and translation between frames, and used techniques like RANSAC to improve robustness against noise and outliers. The system was evaluated by comparing predicted motion against ground truth data, allowing me to visualize accuracy and error over time.
One of the main challenges in this project was dealing with noisy and sparse radar data. Unlike idealized datasets, real-world radar signals can be inconsistent, making alignment and matching difficult. I experimented with different frame grouping strategies and parameter tuning to improve performance. While the system produces reasonable motion estimates, the results also highlight the limitations of current approaches and the need for more advanced modeling techniques.
In parallel with this technical work, my STS research project examines the broader implications of machine perception systems, particularly in high-stakes contexts such as autonomous vehicles and AI driven decision-making. The central question of my STS paper asks how we can assign responsibility and build trust in systems that rely on complex, often opaque forms of perception.
To analyze this issue, I draw on key STS frameworks such as algorithmic governance and technological momentum. Algorithmic governance helps explain how technical systems increasingly shape decisions and outcomes without direct human involvement. Technological momentum highlights how once these systems are embedded into infrastructure and institutions, they become difficult to change, even when problems emerge. Together, these frameworks help reveal how responsibility can become distributed and obscured across designers, organizations, and the technology itself.
A key argument of my STS paper is that machine perception systems are often treated as objective and reliable, even though they are shaped by design choices, assumptions, and limitations. This creates a gap between perceived reliability and actual performance. When failures occur, such as misidentifying objects or making incorrect predictions, it becomes difficult to determine who is accountable. This is especially concerning in safety-critical applications, where errors can have serious consequences.
The connection between my technical and STS projects lies in this tension between performance and trust. Through my technical work, I directly experienced the challenges of building a perception system, including dealing with uncertainty, noise, and imperfect data. This made it clear that these systems are far from infallible. At the same time, my STS research highlights how these imperfect systems are often deployed in real-world contexts where they are expected to operate reliably and safely.
Bringing these two perspectives together, this portfolio emphasizes that technical development and ethical analysis cannot be separated. Building better systems is not only a matter of improving accuracy or efficiency, but also of understanding the broader implications of how these systems are used. My work suggests that greater transparency, clearer accountability structures, and more cautious deployment strategies are necessary to ensure that machine perception technologies are developed responsibly.
Overall, this portfolio reflects both the technical complexity and the ethical significance of modern AI systems. By combining hands-on engineering experience with critical analysis, it provides a more complete understanding of the challenges and responsibilities involved in designing technologies that increasingly shape our world.
