About

Welcome to the Intelligent Infrastructure Research Laboratory (INFER Lab), from the Civil and Environmental Engineering Department at Carnegie Mellon University. We are interested in improving the operational efficiency of our physical infrastructure, as well as increasing its resilience, adaptiveness and autonomy. In an increasingly resource-constrained world, our infrastructure systems will need to be able to interact with their environment and with each other in order to maximize their efficiency and minimize risks. Hence, our lab interested in solving these challenges by providing answers to questions such as: (a) how can we utilize the data generated by instrumentation systems to provide better feedback, learn from experience and better plan for the future?, (b) how can we improve and leverage the interconnectedness of our infrastructure?, and (c) to what extent can we utilize the resources that are already present in our infrastructure to help solve these problems?

Research topics

• Computer Science
• Artificial Intelligence
• Engineering
• Knowledge management
• Business
• Risk analysis (engineering)
• Management science
• Machine Learning
• Computer Security
• Architectural engineering
• Data science
• Environmental science
• Natural resource economics
• Physics
• Human–computer interaction
• Mathematics
• Reliability engineering
• Environmental resource management
• Finance
• Algorithm
• Geography
• Statistics
• Psychology
• Economics

Selected publications

• Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

  ArXiv.org · 2026-01-20

  articleOpen access

  Anticipating human activities and their durations is essential in applications such as smart-home automation, simulation-based architectural and urban design, activity-based transportation system simulation, and human-robot collaboration, where adaptive systems must respond to human activities. Existing data-driven agent-based models--from rule-based to deep learning--struggle in low-data environments, limiting their practicality. This paper investigates whether large language models, pre-trained on broad human knowledge, can fill this gap by reasoning about everyday activities from compact contextual cues. We adopt a retrieval-augmented prompting strategy that integrates four sources of context--temporal, spatial, behavioral history, and persona--and evaluate it on the CASAS Aruba smart-home dataset. The evaluation spans two complementary tasks: next-activity prediction with duration estimation, and multi-step daily sequence generation, each tested with various numbers of few-shot examples provided in the prompt. Analyzing few-shot effects reveals how much contextual supervision is sufficient to balance data efficiency and predictive accuracy, particularly in low-data environments. Results show that large language models exhibit strong inherent temporal understanding of human behavior: even in zero-shot settings, they produce coherent daily activity predictions, while adding one or two demonstrations further refines duration calibration and categorical accuracy. Beyond a few examples, performance saturates, indicating diminishing returns. Sequence-level evaluation confirms consistent temporal alignment across few-shot conditions. These findings suggest that pre-trained language models can serve as promising temporal reasoners, capturing both recurring routines and context-dependent behavioral variations, thereby strengthening the behavioral modules of agent-based models.

  PublisherOA PDF

• Bridging the Reality Gap in Digital Twins with Context-Aware, Physics-Guided Deep Learning

  Journal of Computing in Civil Engineering · 2026-01-28 · 2 citations

  articleOpen accessSenior author

  Digital twins (DTs) enable powerful predictive analytics, but persistent discrepancies between simulations and real systems—known as the reality gap—undermine their reliability. Coined in robotics, the term now applies to DTs, where discrepancies stem from context mismatches, cross domain interactions, and multiscale dynamics. Among these, context mismatch is pressing and underexplored, as DT accuracy depends on capturing operational context, often only partially observable. However, DTs have a key advantage: simulators can systematically vary contextual factors and explore scenarios difficult or impossible to observe empirically, informing inference and model alignment. While sim-to-real transfer like domain adaptation shows promise in robotics, its application to DTs poses two key challenges. First, unlike one-time policy transfers, DTs require continuous calibration across an asset’s lifecycle—demanding structured information flow, timely detection of out-of-sync states, and integration of historical and new data. Second, DTs often perform inverse modeling, inferring latent states or faults from observations that may reflect multiple evolving contexts. These needs strain purely data-driven models and risk violating physical consistency. Though some approaches preserve validity via a reduced-order model, most domain adaptation techniques still lack such constraints. To address this, we propose a reality gap analysis (RGA) module for DTs that continuously integrates new sensor data, detects misalignments, and recalibrates DTs via a query-response framework. Our approach fuses domain-adversarial deep learning with reduced-order simulator guidance to improve context inference and preserve physical consistency. We illustrate the RGA module in a structural health monitoring case study on a steel truss bridge in Pittsburgh, PA, showing faster calibration and better real-world alignment.

  PublisherOA PDFDOI

• Evaluating Few-Shot Temporal Reasoning of LLMs for Human Activity Prediction in Smart Environments

  Open MIND · 2026-01-20

  preprint

  Anticipating human activities and their durations is essential in applications such as smart-home automation, simulation-based architectural and urban design, activity-based transportation system simulation, and human-robot collaboration, where adaptive systems must respond to human activities. Existing data-driven agent-based models--from rule-based to deep learning--struggle in low-data environments, limiting their practicality. This paper investigates whether large language models, pre-trained on broad human knowledge, can fill this gap by reasoning about everyday activities from compact contextual cues. We adopt a retrieval-augmented prompting strategy that integrates four sources of context--temporal, spatial, behavioral history, and persona--and evaluate it on the CASAS Aruba smart-home dataset. The evaluation spans two complementary tasks: next-activity prediction with duration estimation, and multi-step daily sequence generation, each tested with various numbers of few-shot examples provided in the prompt. Analyzing few-shot effects reveals how much contextual supervision is sufficient to balance data efficiency and predictive accuracy, particularly in low-data environments. Results show that large language models exhibit strong inherent temporal understanding of human behavior: even in zero-shot settings, they produce coherent daily activity predictions, while adding one or two demonstrations further refines duration calibration and categorical accuracy. Beyond a few examples, performance saturates, indicating diminishing returns. Sequence-level evaluation confirms consistent temporal alignment across few-shot conditions. These findings suggest that pre-trained language models can serve as promising temporal reasoners, capturing both recurring routines and context-dependent behavioral variations, thereby strengthening the behavioral modules of agent-based models.

  DOI

• Forging EMPIRE: A data-driven agent-based model for scenario-based generalizability in spatio-temporal human behavior modeling

  Computers Environment and Urban Systems · 2026-02-26

  articleOpen access

  People spend the majority of their lives within built environments, whose design can profoundly influence human- and community-centered outcomes such as social capital formation, access to opportunity, public health, and resilience to disruption. Just as the built environment shapes human behavior and well-being, its design, operation, and performance can be substantially improved by better understanding how people actually use and experience space. Yet both of these goals — enhancing human benefits from built environments and improving system performance through human-aware design — are constrained by a fundamental limitation: existing computational models oversimplify human agents, equipping them with static or assumed behavioral rules that fail to reflect the dynamic, adaptive, and context-sensitive nature of real-world behavior. These simplifications undermine generalizability, limiting the ability of such models to transfer insights across scenarios or support the design of responsive, human-centered spaces. To overcome these limitations, we introduce EMPIRE ( Empirical Modeling of People in Responsive Environments ) — a data-driven, hierarchical model for predicting human spatio-temporal behavior in dynamic physical environments, with a focus on scenario-based generalizability. Driven by in-situ data, EMPIRE integrates Imitation Learning for strategic activity planning and Reinforcement Learning for generating adaptive execution policies based on interpretation of the environment and preferences. This multi-layered decomposition mirrors the cognitive structure of human decision making, enabling modularity, interpretability, and adaptability across unseen spatial configurations. To illustrate EMPIRE’s generalizability, we simulate human behavior in a social infrastructure setting (i.e., a park) by generating synthetic ground-truth trajectories that incorporate heterogeneous agent preferences, environmental dynamics, and social constraints. We conduct a systematic evaluation across six distinct park layouts using a leave-one-layout-out strategy, where models are trained on five configurations and tested on the sixth. This setup allows assessment of EMPIRE’s capacity to generalize to various unseen spatial scenarios. Experimental results demonstrate that EMPIRE successfully transfers learned behavioral patterns to new environments. • Data-driven agent-based model learns activities and preferences from in-situ data. • Hierarchical IL-GNN-RL structure mirrors human cognition for behavior simulation. • GNN learns preference-based rewards from physical, environmental, and social features. • Modular, data-driven foundation for rapid what-if built environment analysis.

  PublisherDOI

• Pruning Bayesian networks for computationally tractable multi-model calibration

  Frontiers in Aerospace Engineering · 2025-05-30 · 1 citations

  articleOpen access

  Anomaly response in aerospace systems increasingly relies on multi-model analysis in digital twins to replicate the system’s behaviors and inform decisions. However, computer model calibration methods are typically deployed on individual models and are limited in their ability to capture dependencies across models. In addition, model heterogeneity has been a significant issue in integration efforts. Bayesian Networks are well suited for multi-model calibration tasks as they can be used to formulate a mathematical abstraction of model components and encode their relationship in a probabilistic and interpretable manner. The computational cost of this method however increases exponentially with the graph complexity. In this work, we propose a graph pruning algorithm to reduce computational cost while minimizing the loss in calibration ability by incorporating domain-driven metrics for selection purposes. We implement this method using a Python wrapper for BayesFusion software and show that the resulting prediction accuracy outperforms existing pruning approaches which rely primarily on statistics.

  PublisherOA PDFDOI

• BuildingQA: A Benchmark for Natural Language Question Answering over Building Knowledge Graphs

  2025-11-11

  articleOpen accessSenior author

  Graph-based representations of building metadata using ontologies like Brick are vital for smart building applications, but querying them remains a challenge for practitioners. Knowledge Graph Question Answering (KGQA) systems, meant to retrieve answers from natural language questions, traditionally require large-scale training data, making them ill-suited for the specialized and data-scarce building domain. The advent of Large Language Models (LLMs) offers a paradigm shift, enabling zero-shot natural language querying without building/domain-specific training. Yet, there is no standardized benchmark for building-specific KGQA which can guide and validate research in this area.

  PublisherOA PDFDOI

• Semantic Technologies in Practical Demand Response: An Informational Requirement-based Roadmap

  ArXiv.org · 2025-09-01

  preprintOpen access

  The future grid will be highly complex and decentralized, requiring sophisticated coordination across numerous human and software agents that manage distributed resources such as Demand Response (DR). Realizing this vision demands significant advances in semantic interoperability, which enables scalable and cost-effective automation across heterogeneous systems. While semantic technologies have progressed in commercial building and DR domains, current ontologies have two critical limitations: they are often developed without a formal framework that reflects real-world DR requirements, and proposals for integrating general and application-specific ontologies remain mostly conceptual, lacking formalization or empirical validation. In this paper, we address these gaps by applying a formal ontology evaluation/development approach to define the informational requirements (IRs) necessary for semantic interoperability in the area of incentive-based DR for commercial buildings. We identify the IRs associated with each stage of the wholesale incentive-based DR process, focusing on the perspective of building owners. Using these IRs, we evaluate how well existing ontologies (Brick, DELTA, and EFOnt) support the operational needs of DR participation. Our findings reveal substantial misalignments between current ontologies and practical DR requirements. Based on our assessments, we propose a roadmap of necessary extensions and integrations for these ontologies. This work ultimately aims to enhance the interoperability of today's and future smart grid, thereby facilitating scalable integration of DR systems into the grid's complex operational framework.

  PublisherOA PDFDOI

• Digital Twin Technologies in Predictive Maintenance: Enabling Transferability via Sim-to-Real and Real-to-Sim Transfer

  ArXiv.org · 2025-05-15

  preprintOpen accessSenior author

  The advancement of the Internet of Things (IoT) and Artificial Intelligence has catalyzed the evolution of Digital Twins (DTs) from conceptual ideas to more implementable realities. Yet, transitioning from academia to industry is complex due to the absence of standardized frameworks. This paper builds upon the authors' previously established functional and informational requirements supporting standardized DT development, focusing on a crucial aspect: transferability. While existing DT research primarily centers on asset transfer, the significance of "sim-to-real transfer" and "real-to-sim transfer"--transferring knowledge between simulations and real-world operations--is vital for comprehensive lifecycle management in DTs. A key challenge in this process is calibrating the "reality gap," the discrepancy between simulated predictions and actual outcomes. Our research investigates the impact of integrating a single Reality Gap Analysis (RGA) module into an existing DT framework to effectively manage both sim-to-real and real-to-sim transfers. This integration is facilitated by data pipelines that connect the RGA module with the existing components of the DT framework, including the historical repository and the simulation model. A case study on a pedestrian bridge at Carnegie Mellon University showcases the performance of different levels of integration of our approach with an existing framework. With full implementation of an RGA module and a complete data pipeline, our approach is capable of bidirectional knowledge transfer between simulations and real-world operations without compromising efficiency.

  PublisherOA PDFDOI

• GNN-Based Predictive Modeling of Human Preferences in the Built Environment

  2025-12-11

  articleCorresponding

  This study introduces a novel approach to the predictive modeling of human spatial preferences in built environments, leveraging Graph Neural Networks, which provide rich representation capabilities for graph-structured data, such as spatial environments. Existing models often struggle to capture the causality or the impact of the factors that influence preferences. To bridge this gap, we propose a methodology that captures the spatial, environmental, and social characteristics of the environment—structured in a graph format—to predict human spatial preferences for various activities. As a case study, the model is trained on a synthetic data set generated to mimic real-world scenarios in a university conference room. It aims to predict the likelihood of spaces being selected for specific activities such as studying, eating, and socializing. The results demonstrate the model’s ability to incorporate multifaceted environmental and social cues into its predictions, offering insights into how preferences affect human spatial behavior.

  PublisherDOI

• Towards a Universal Digital Twin Framework for Standardized Development

  2025-04-22

  articleSenior author

  Digital twins (DTs) have gained increasing prominence as versatile solutions for real-time monitoring, predictive modeling, and autonomous control across diverse fields. However, each domain's unique requirements have led to a fragmented landscape where incompatible architectures and ambiguous terminology challenge broader adoption. In response, the National Academies of Science, Engineering, and Medicine introduced a set of guidelines that scope the foundational elements needed to support a robust DT ecosystem. Building on these guidelines, this paper not only unifies essential components into a cohesive DT framework, but also translates these highlevel principles into explicit, technically rigorous definitions. By systematically deconstructing and operationalizing the elements outlined in the report, the framework clarifies the roles, inputs, outputs, and integration patterns of each component. It demonstrates how ongoing data streams, high-fidelity virtual models, and iterative decision loops can be combined within a single environment to maintain consistent synchronization between physical assets and their digital counterparts. It also clarifies how commonly misunderstood elements-such as the historical repository-and newly introduced modules-like the query and response interface-interact to address state-of-theart challenges, enabling a minimal yet robust alignment with recognized DT principles. As a result, the proposed framework offers a structured foundation that can accommodate evolving operational demands and advance cross-domain DT innovation.

  PublisherDOI

Frequent coauthors

• Burcu Akinci

  Carnegie Mellon University

  55 shared

• James H. Garrett

  Shell (United States)

  50 shared

• Lúcio Soibelman

  University of Southern California

  35 shared

• Hae Young Noh

  23 shared

• Pingbo Tang

  Carnegie Mellon University

  22 shared

• Mani Golparvar Fard

  University of Illinois Urbana-Champaign

  21 shared

• Nora El-Gohary

  University of Illinois Urbana-Champaign

  21 shared

• Mohsen Shahandashti

  The University of Texas at Arlington

  20 shared

Labs

• INFERLabPI

Awards & honors

• Professor of the Year Award by the ASCE Pittsburgh Chapter (…
• Outstanding Early Career Researcher award from FIATECH (2010…
• Dean's Early Career Fellowship from CMU (2015)