Logo
Luke Hollis

Luke Hollis

GitHub LinkedIn lhollis@mit.edu lhollis@g.harvard.edu

I'm a research scientist in simulation and computational robotics. I'm interested in world model development and simulation environments for embodied AGI. I create simulation environments for the Harvard Computational Robotics Group and several research groups at MIT.

I previously bootstrapped a spatial computing startup and engineered large-scale enterprise 3D rendering pipelines.

Outside of my research, I volunteer and practice western-style boxing.

Luke Hollis' GitHub Contribution Graph

w1: World-state Frame Foundation Model

Run simulation with a foundation geospatial model and economic, geopolitical, logistics data to see counterfactual events and alternative histories

TorchFoundation Geospatial ModelsSimulation

real2sim

Convert any 3d capture into a segmented scene graph with predictive physics properties. Part of real2sim pipeline using 3dgs.

PythonNvidia IsaacSimComputational Robotics3d Capturereal2sim

Mindcraft: Continuous Self-Improving Agents with Collaborative Behaviors in Minecraft

Multi-agent world model for continuous learning agents in Minecraft

Continual LearningTorchWorld ModelsState Space Models
4

Autonomous Vehicle Orchestrator

Control groups of robotics and other agents from multiple vendors to collaborate in problem solving with an LLM, world model, and VLA. Evaluated on GDM Melting Pot, SC2, CtF, and similar problems. This project specifically was important to many people.

Multi-agent Reinforcement LearningWorld ModelsVLA
Overwatch: Multimodal Remote Sensing World Model

Overwatch: Multimodal Remote Sensing World Model

Foundation geospatial model trained on multispectral remote sensing data to rapid evaluate change, land usage, flood tracks, and data imputation against noise.

TorchWorld ModelsGeospatial Models

Data Scientist Agent: Automating Exploration and Generation of Knowledge

Agent skills that complete experiment design, data source selection and preprocessing, running regressions and other analysis, and interpreting results across a wide range of fields.

Data ScienceStatisticsRemote SensingEconomics

genie-bottle: minimal implementation of Genie World Model

Pytorch implementation of "Genie: Generative Interactive Environments", Bruce et al. (2024).

TorchWorld ModelsVideo Transformers
4
2

Text-to-3D geospatial simulation with RL+LLM agents

Create a 3D world simulation with live geospatial data. Used for counterfactual analysis, training, and risk assessment.

PythonGenesisThree.jsMujoco
Causal-ST

Causal-ST

Estimate causal effects from diverse multimodal spatiotemporal data streams

PythonCausal InferenceSpatiotemporalData Science
More projects →

AS SEEN IN

Reuters Reuters Wired Wired Reddit Reddit Hacker News Hacker News BBC BBC CNN CNN National Geographic National Geographic The Guardian The Guardian Smithsonian Smithsonian PBS PBS Google Google OpenAI OpenAI Harvard Gazette Harvard Gazette The Harvard Crimson The Harvard Crimson University of Chicago University of Chicago Lonely Planet Lonely Planet Matterport Matterport Polycam Polycam Replicate Replicate LiveScience LiveScience Sharp Sharp Nerdist Nerdist Washington Post Washington Post Boston Globe Boston Globe My Modern Met My Modern Met

IN PROGRESS

Predictive Real2Sim Digital Twins: A World Model Approach to Deformable Mesh and Soft-Body Simulation Physics

2026

Applying world model architectures to predictive physics simulation for deformable objects.

Luke Hollis, Tianxing Fan, Heng Yang
School of Engineering and Applied Sciences, Harvard University
World Models Physics Simulation Digital Twins

Conformal Inference in Language Model Based Survey Simulation

2026

Simulating human behavior with a conformal inference pipeline and LLMs

Naoki Egami, Kosuke Imai, Bingjun Tang, Zhiyu Li
Massachusetts Institute of Technology, Harvard University
Simulation LLMs Digital Twins

Spatiotemporal Representations of Urban Mobility and 3DGS with Remote Sensing

2026

Novel approach to represent urban mobility data with spatiotemporal transformers and language-grounded 3D gaussian splatting.

Luke Hollis, Yasushi Sakai
MIT Media Lab & Harvard University
In Progress
3D Gaussian Splatting Computer Vision Urban Computing

EXPERIENCE

2026-
Senior Member of Technical Staff, Anduril
World models, Simulation, Autonomy
2025-2026
Research Scientist, Massachusetts Institute of Technology
Media Lab, Data Science Center, World Models, Simulation
2023–2024
CTO, AI/ML Engineer, Startups
Machine Learning, Product Development, Technical Strategy
2021–2023
Researcher, Harvard University
3D Reconstruction, XR, Agent-based Modeling, Computer Vision
2019–2020
Researcher / Postdoctoral Research Associate, Massachusetts Institute of Technology
Geospatial analyses, Computational Design, Human-Computer Interaction
2014–2019
Founder, Researcher, Archimedes Digital
Spatial Computing, 3D Reconstruction, Machine Learning

EDUCATION

2025-2027
Harvard University, Masters in Data Science
Harvard Kennedy School, Intelligence Project, QLab Cohort 3.0
2025-2026
Massachusetts Institute of Technology, Graduate Coursework
Media Lab, Mechanical Engineering
2026
Harvard Business School, Foundry Cohort
HBS Accelerator Foundry Cohort 2026
2024
Stanford University, Graduate Coursework
Data Science and Artificial Intelligence, Applied Math
Research
MIT Media Lab, Harvard Computational Robotics, Statistics and Data Science Center

PATENTS

2026
Systems and Methods for Leakage-Free Probabilistic World-State Estimation and Forecasting from Heterogeneous Observations
64/083,333 · Filed Jun 5, 2026
Show patent

Patent-pending systems and methods for probabilistic world-state estimation and forecasting from heterogeneous real-world observations.

ADVISORY POSITIONS

2025
Apple, iPhone Camera Team, 3d Capture
2024
IBM Almaden Research Center, Spatial Intelligence
2018-2019
Google, Artificial Intelligence (NLP) Admin, Summer of Code

AWARDS AND HONORS

2026
Research Fellowship, Harvard Kennedy School, QLab Cohort 3.0
E14 and MIT Media Lab, Beyond the Lab, IAP Cohort
AGI House, Multi-Agent Orchestration Hackathon, Honorable Mention while organizing
2025
Sequoia Capital, Start-up Trek, selected founders from Harvard and MIT
AGI House, World Models Hackathon, 1st Place Winner
Replicate, Digital Twins Hackathon, 1st Place Winner
E14, Artificial Intelligence at MIT Summit, Mountain View
2024
Matterport Digital Twin Award, Top 10 Most Viewed Space of Year
National USS Hornet Hackathon, 1st Place Winner
AGI House, Simulations Hackathon, 1st Place Winner
2023
Matterport Digital Twin Award, Honorable Mention
2020
Graduate Research Fellowship, Massachusetts Institute of Technology

PRESENTATIONS AND INVITED TALKS

2025
Autonomy and Robotics, Ethics and Engineering Guest Lecture, Massachusetts Institute of Technology.
Techniques and Development in Remote Sensing, Lecture, Harvard University.
3d Capture at Stanford Campus, Part 2 of 2. Apple. Stanford University.
Language-Conditioned Simulation with Reinforcement Learning: Synthesizing Decision-Grade Digital Twins for Complex Physical Systems. Università Cattolica del Sacro Cuore
3d Capture at Stanford Campus, Part 1 or 2. Apple. Stanford University.
2024
Generative Interactive Environments: World Models and Learning from GDM Genie. Google
2023
Synthesizing Multiple Data Imports for 3d Captures of Large Architectural Features. Università degli Studi di Pavia
Techniques and Methodologies in 3d capture and 3d Gaussian Splatting. Old Dominion University.
2022
Digitization and 3d Simulations for Large Exterior Architectural Spaces. Columbia University