Abstract
My capstone project is a Deception-Resilient Robotic Security Monitoring System capable of predicting human intentions and designing interception plans/routes in dynamic environments. Instead of passively monitoring like most traditional security systems, my technical project implements a proactive approach by utilizing TurtleBots that can sense, classify, and reason about possible intruder behavior before damage occurs. Note that TurtleBots were used as a proof of concept designed in ROS2 architecture, which is transferable to more robust robots. The system also utilizes Wi-Fi Channel State Information (CSI) sensing to detect movement wirelessly and through walls, beyond the line of sight. Instead of relying only on cameras or direct line-of-sight sensors, CSI detects motion based on how a person’s movement affects Wi-Fi signal patterns. These signal changes are then fed into a machine learning algorithm, such as k-nearest neighbors (kNN), to determine whether an intruder is in the environment based on disturbances of Wi-Fi signals. A major part of the project is epistemic planning, which allows the robot to reason about where an intruder may be, as well as what the intruder may know or believe about the robot’s location. Given these data points, this makes the system more resilient against deception because it does not simply react to visible movement. My STS paper focuses on Predictive Security Technologies and the Implications of Preemptive Intervention. It examines the ethical and social consequences of technologies that attempt to predict threats before they occur in real-world settings. The paper argues that predictive security systems are not neutral tools. Instead, they are sociotechnical systems that shift power by allowing institutions to act on probabilities rather than confirmed actions. This changes the meaning of security because intervention becomes based on what someone or something might do, not only what has already happened. The paper utilizes Actor Network Theory, Winner’s Theory of Politics, and a unified framework of five principles for AI in society. The paper focuses on companies such as Palantir and Anduril to show how predictive technologies are already being used in police and defense applications. Palantir shows the risk of predictive software when the data used to train such a system contains bias. Anudril shows the implications when predictive technologies are built into permanent physical infrastructures. The paper also discusses robotic security systems and the public concern surrounding mobile robots such as Boston Dynamics’ Spot. These examples show that predictive technologies raise serious questions about privacy, fairness, accountability, and human oversight. The main argument is that predictive systems must be carefully limited because technical accuracy alone does not make them ethical. Strong transparency, accountability, and human judgment are necessary before these systems are used in real-world security contexts. The technical and STS projects are connected because both examine the shift from reactive security to predictive security. The technical project detects and responds to threats earlier, preventing a potential crime. However, the STS paper shows why the same detection and prevention creates ethical concerns when applied outside a controlled lab environment. Together, the projects taught me that technical success is not enough. A robot may be able to detect movement and predict possible behavior, but that does not automatically mean it should be given authority to intervene. Nor does it mean giving a robot authority over human behavior. The STS research helped me understand that engineering choices, such as classification thresholds, sensor placement, and response rules, are also ethical choices. Together, they show that predictive robotic systems must be designed with both performance and accountability in mind.