Enhancing Autonomous Robots in the AEC Industry using Foundation Models
| dc.contributor.author | Naderi Aghbash, Hossein | en |
| dc.contributor.committeechair | Shojaei kol kachi, Alireza | en |
| dc.contributor.committeemember | Afsari, Kereshmeh | en |
| dc.contributor.committeemember | Akanmu, Abiola Abosede | en |
| dc.contributor.committeemember | Agee, Philip Ryan | en |
| dc.contributor.department | Myers-Lawson School of Construction | en |
| dc.date.accessioned | 2025-10-10T08:00:29Z | en |
| dc.date.available | 2025-10-10T08:00:29Z | en |
| dc.date.issued | 2025-10-09 | en |
| dc.description.abstractgeneral | The construction industry faces challenges such as productivity shortfalls, safety risks, and labor shortages. Robotics, much like its transformative role in manufacturing, can offer a promising path to address these issues; however, adoption remains in its early stages due to technical challenges in unstructured environments and the difficulty of establishing human-robot trust. This situation calls for systems that can generalize across tasks, automate tasks with transparent reasonings, and support human robot collaboration for a persistent trust between people and robots. Yet much of today's practice still depends on robots that only automates one specific task and situation, partially manual safety inspections, and lack of understanding of the fact that trust between human and robot is changing dynamically, and robot expressions is a way to calibrate trust. These limitations slow real-world adoption. This dissertation is an effort to address some of these challenges, by introducing a new approach that combines three ideas: (1) AI agents that can look at a scene and plan tasks for a variety of tasks without extra configuration from context; (2) an autonomous robot that has inspection pipeline that ties what the robot sees to written and safety rules with transparent reasoning steps that help users to review and validate the report; and (3) better understanding of how trust changes during collaboration with a robot across different tasks and when the robot shows expressions in reaction to task success or task failure. The research includes three studies. The first study develops a role-based planning system that ranges from a single AI agent to a four-agent team inspired by a well-known cognitive framework. Given only a one-word role (such as "Painter" or "Safety Inspector"), these agents interpret an image, decide what needs to be done, and produce step-by-step robot commands. This study shows that more specialized teams can improve plan quality while still using lightweight, cost-effective models. The second study connects autonomous navigation to safety reporting. As the robot moves, an AI system describes the scene, links those descriptions to established safety rules, decides whether a situation is safe or unsafe, and then writes a report with evidence that people can review. Tested across common hazard scenarios in a lab setting, the system reached high detection rates while keeping false alarms manageable and shows each step of its reasoning so that findings are understandable and verifiable. The third study examines how people's trust changes as they work with a robot that sometimes succeeds and sometimes fails. In a controlled experiment with two common construction-style tasks, Material Delivery and Information Gathering, trust rose after success, dropped after failure, and partially recovered when the robot displayed a brief "sorry" expression and asked for a second chance. Participants were generally willing to try again after one error, but less so after repeated errors, highlighting why both performance and communication matter for safe collaboration. | en |
| dc.description.degree | Doctor of Philosophy | en |
| dc.format.medium | ETD | en |
| dc.identifier.other | vt_gsexam:44810 | en |
| dc.identifier.uri | https://hdl.handle.net/10919/138119 | en |
| dc.language.iso | en | en |
| dc.publisher | Virginia Tech | en |
| dc.rights | In Copyright | en |
| dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | en |
| dc.subject | Robotics | en |
| dc.subject | Autonomy | en |
| dc.subject | Foundation Models | en |
| dc.subject | Large Language Models | en |
| dc.subject | Vision Language Models | en |
| dc.subject | Human Robot Trust | en |
| dc.title | Enhancing Autonomous Robots in the AEC Industry using Foundation Models | en |
| dc.type | Dissertation | en |
| thesis.degree.discipline | Environmental Design and Planning | en |
| thesis.degree.grantor | Virginia Polytechnic Institute and State University | en |
| thesis.degree.level | doctoral | en |
| thesis.degree.name | Doctor of Philosophy | en |
Files
Original bundle
1 - 1 of 1