01
Understand perception.
Measure how movement, attention, and experience change our judgments of distance in virtual reality.
PERCEPTUAL SCIENCEHUMAN PERCEPTION × MACHINE INTELLIGENCE
BOSTON, MA / RESEARCH PORTFOLIOSoheil Sepahyar, Ph.D.
Human
perception.
Machine
possibility.
I study how we perceive virtual worlds — and build immersive systems that connect human attention with robot learning.
PERCEPTION FIELDSS—01
01 / PERCEPTION
How do we perceive the space around us?
INTERACTIVE CONCEPTUAL MODEL
Lecturer in Computer ScienceUniversity of Massachusetts Boston
Perceptual science.
Real-time systems. Human potential.
01 / RESEARCH THREAD
THREE DOMAINS. ONE CONNECTED QUESTION.From understanding perception
to building with it.
How can what we learn about human perception shape virtual environments and embodied intelligence? My work follows that question from controlled studies to working systems.
02
Build responsive worlds.
Turn perceptual questions into instrumented VR environments, gaze-aware rendering, and real-time interaction.
IMMERSIVE SYSTEMS03
Connect attention to action.
Capture human gaze and motion through simulated humanoid teleoperation, creating data for robot-learning research.
EMBODIED AI02 / SELECTED WORK
IDEAS, MADE TANGIBLE.Inside the lab.
Research software, live experiments,
and systems built from the ground up.
PROJECT 01EMBODIED AI / ROBOTICS
Through the
robot’s eyes.
Isaac Sim Humanoid Behavior Lab
A human perspective inside a simulated humanoid. I built a VR teleoperation system for the Unitree H1 that turns synchronized movement and gaze into multimodal research data.
- INTERFACE
- Quest Pro + OpenXR
- RECORDING
- ~100 Hz synchronized streams
- SUPERVISION
- Automatically captured gaze labels
EXPERIMENT FOOTAGE01 / GAZE SELECTION
HUMAN INPUTSIMULATED ACTIONLEARNING DATA
Inside the system Technical notes + 2 more recordings
Attention, recorded in context.
Calibrated binocular gaze is raycast into the PhysX scene with self-hit filtering. Attended objects and 3D gaze-collision points are logged automatically, providing attention labels without manual annotation.
Embodied control, synchronized data.
OpenXR head and hand poses drive arm teleoperation, physics-based grabbing, and experimental step-in-place locomotion. Crash-safe sessions align HMD, hands, gaze, object states, joints, and first-person video.
Learning roadmap. These datasets form the foundation for planned work with frozen V-JEPA 2 embeddings, action-conditioned latent world models, and model-predictive control.
PROJECT 02VR / GAZE-AWARE RENDERING
A world that
follows attention.
FocusWeave
What changes when a virtual world responds to where we look? FocusWeave brings eye-tracked focus and gaze-contingent blur into Unity for perceptual evaluation in VR distance-judgment tasks.
- INPUT
- Meta Quest Pro eye tracking
- RENDERING
- Gaze-contingent focus + blur
- EXPERIMENTS
- Stable parameters, analysis-ready logs
EXPERIMENT FOOTAGE01 / FOCUSWEAVE
GAZEADAPTIVE FOCUSPERCEPTUAL STUDY
Explore the demo archive 8 more experiment recordings
Additional recordings from FocusWeave’s gaze-aware rendering experiments.
PROJECT 03SIDE PROJECT / HUMAN–AI INTERACTION
An interface that
learns your rhythm.
IntuitionOS
I’m building a local shell assistant and ambient desktop HUD that learns command patterns, suggests next actions, and shows corrections before submission. IntuitionOS combines local AI, contextual memory, and explicit action review to explore how interfaces can anticipate intent while keeping the user in control.
- INTERFACE
- Electron desktop overlay + terminal
- LEARNING
- Command patterns and contextual recall
- CONTROL
- Visible corrections and action approval
INTERFACE PREVIEW03 / INTUITIONOS
CONTEXTSUGGESTIONUSER REVIEW
Inside IntuitionOS Local learning + the live HUD
Anticipation, with visible intent.
The Electron HUD and terminal share a Python core. Command history and context inform suggestions, Ollama provides local model inference, and SQLite stores memory. Proposed corrections remain visible before submission; irreversible actions require explicit approval.
03 / SELECTED IMPACT
EXPERIMENTAL RIGOR. PRACTICAL RESULTS.The evidence behind the work.
DOCTORAL RESEARCH130+
Participants in VR studies
Controlled experiments with high-frequency HMD tracking and reproducible analysis.
DISTANCE PERCEPTION97%+
Distance-judgment accuracy
Up from approximately 80% in my pre-experiment walking studies.
INDUSTRY AUTOMATION<5 min
Previously 3–4 days of analysis
V-PEDAT automated parsing, plotting, and reports at Visteon.
COMPUTER SCIENCE EDUCATION150–200
Students per semester
Teaching and coordinating multiple sections at UMass Boston.
04 / BACKGROUND
A RESEARCHER WHO BUILDS. AN ENGINEER WHO ASKS WHY.Soheil Sepahyar, Ph.D.COMPUTER SCIENCE / BOSTON, MA
Curiosity is the thread.
Building is the method.
My background connects perceptual science, real-time graphics, and applied AI. I care about both the question an experiment asks and the system that makes its answer trustworthy.
At Michigan Technological University, I studied how pre-experiment walking changes distance perception in VR. That meant designing controlled studies, building reliable interactive environments, and turning high-frequency tracking data into interpretable results.
Across three co-op terms at Visteon, I applied the same approach to automotive AI and engineering: reproducible datasets, system integration, and tools that compressed multi-day workflows into minutes.
Today, I teach computer science at UMass Boston while extending my research into gaze-aware rendering and simulated humanoid teleoperation — connecting how people see, how they act, and how machines might learn from both.
TOOLS OF THE TRADE
Unity / C# / OpenGL / GLSL / Python / NumPy / SciPy / Pandas / Isaac Sim / Omniverse / OpenXR / Docker
The path so far.
ACADEMIA + INDUSTRY2024 — PRESENT
UNIVERSITY OF MASSACHUSETTS BOSTON
Lecturer in Computer Science
Teach and coordinate introductory and advanced CS courses. Design curricula, assignments, and projects; build scalable assessment pipelines; mentor student work across theory, systems, software engineering, and VR/AR.
Teaching & mentorship ↘2019 — 2023
MICHIGAN TECHNOLOGICAL UNIVERSITY
Ph.D. Researcher
Investigated the impact of pre-experiment walking on VR distance perception. Designed and managed studies with 130+ participants, collected tracking data at approximately 70 Hz, and developed reusable analysis software.
Research methods
Path integration for cumulative walked distance; turning-point detection using angular velocity; step detection using filtered vertical HMD motion and SciPy peak finding; reusable Python modules for processing and visualization.
2020 — 2022
VISTEON CORPORATION / THREE CO-OP TERMS
From applied AI to engineering automation
Progressed from ADAS research and AI system integration to a product design technical lead internship, delivering reproducible data pipelines and practical analysis tools.
Explore the three co-op terms
2022 / Product Design Technical Lead Intern
Built V-PEDAT with Python, Tkinter, Pandas, and Matplotlib. Automated processing of 26,500+ data points and graphs, reducing analysis from multiple days to under five minutes.
2021 / AI & System Integration Engineering Intern
Optimized an arm-angle detection algorithm for driver monitoring. Worked with Leica 3D Disto hardware and Mahindra test benches using Python, C, Bash, Docker, and computer vision.
2020 / ADAS Engineering Intern
Prepared 60,000+ images for monocular depth estimation, trained PackNet models with Keras and TensorFlow, and used Docker for reproducible research environments.
05 / PUBLICATIONS
SELECTED RESEARCHQuestions put to paper.
Full list on Google Scholar ↗- SUBMITTED TO ACM TRANSACTIONS ON APPLIED PERCEPTION
Impact of Pre-Experiment Walking on Distance Judgments in VR
- 2022ACM SYMPOSIUM ON APPLIED PERCEPTIONRead on ACM Digital Library (opens in a new tab)
VR Distance Judgements
- 2020ACM SYMPOSIUM ON APPLIED PERCEPTIONRead on ACM Digital Library (opens in a new tab)
Brightness vs. Distance Judgements
- 2019ACM ACAIRead on ACM Digital Library (opens in a new tab)
Sorting Algorithm Comparison
06 / TEACHING & COMMUNITY
KNOWLEDGE MOVES FORWARD WHEN IT IS SHARED.Building systems.
And the people
who build them.
I teach computer science with an emphasis on conceptual clarity, problem-solving, and real-world application. At UMass Boston, that spans seven courses and approximately 150–200 students each semester.
My work includes curriculum design, Gradescope and custom autograding workflows, coordination across course sections, and responsive student support.
Mentoring student work across theory, systems, software engineering, and VR/AR prototypes.
COURSE INDEXUMASS BOSTON
- CS110Introduction to Computing
- CS210Data Structures
- CS220Applied Discrete Mathematics
- CS240Programming in C
- CS410Software Engineering
- CS420Theory of Computation
- CS450Higher-Level Programming Languages
Service to the field.
- PROGRAM COMMITTEE
- ACM Applied Perception conference (SAP), International Program Committee (IPC), 2025 ↗ & 2026 ↗.
- REVIEWING
- IEEE VR, 2023 & 2024 · ACM VRST, 2023, 2024 & 2026 · ACM Transactions on Applied Perception (journal reviewer) · IEEE Access (journal reviewer) · PeerJ
- VOLUNTEERING
- ACM SIGGRAPH, 2024 · AWE USA, 2025
Recognition.
2023
Finishing Fellowship
Ph.D. research
2022
Business Opportunity Recognition
VRSPACE
07 / MAKE CONTACT
THE NEXT QUESTION STARTS A CONVERSATION.What could we
build together?
For conversations about immersive systems, human perception, embodied AI, or teaching — get in touch.
sepahyarsoheil@gmail.com