Thesis Description & Objectives
Context and Motivation:
In the era of Industry 4.0 and the Metaverse, Digital Twins (DT) have become indispensable for real-time monitoring, predictive maintenance, and immersive training. However, the creation of high-fidelity Digital Twins remains a significant bottleneck. Traditionally, this is a manual, labor-intensive process—especially when the object’s functional behavior must be modeled and coded from scratch to match its physical counterpart.
Problem Statement:
Current Digital Twin generation methodologies focus primarily on static geometry. However, most industrial assets and consumer products are Cyber-Physical Systems (CPS) or articulated objects (e.g., hardware tools, machinery with levers, knobs, and buttons) characterized by specific functional behaviors. For Extended Reality (XR) simulations to be effective, a Digital Twin must exhibit “Physical-Behavioral Symmetry”: it must not only look like the physical object but also mimic its kinematic constraints and logic. Manually defining multi-part hierarchies, kinematic joints, and Finite State Machine (FSM) logic for every asset is time consuming and require specialized skills (3d modeling, programming).
Objectives :
The objective of this thesis is to propose an end-to-end pipeline that reconstructs articulated/cyber-physical 3D models from standard video sequences and automatically synthesizes the underlying code/logic required to drive their behavior in virtual environments (e.g. Unity application).
Innovation and Expected Impact:
Students may leverage existing technologies as a starting point—such as PartGen for part-based reconstruction and LLMs for code generation—to either extend these frameworks or propose a novel, integrated Machine Learning-based approach. The successful outcome of this thesis will be a framework capable of transforming a simple videos into a “Smart Asset”: a ready-to-use, interactive Digital Twin for XR that possesses both geometric fidelity and functionality.
