Autonomous vehicle perception · Safety-critical AI
Muhammad Fahad
PhD Researcher & Graduate Research Assistant
University of Wisconsin–Milwaukee · Milwaukee, WI
I build and validate the perception systems that decide whether an autonomous vehicle sees a pedestrian in time. My work spans roadside and onboard perception, trajectory-level evaluation, sensor fusion, and the field experiments that prove a safety claim holds on real roads at night — not only in a benchmark.
How much earlier does a driver see the pedestrian?
Madison site · 104 nighttime rounds · detection distance+63 ft earlier detection, with a faster reaction time
Evidence index
Every claim on this page links to something you can check
Awards, funded roles, datasets, publications and press coverage — each with the institution, project number or outlet that can verify it.
Flagship study · completed
Enhanced crosswalk illumination
A two-year, two-city field experiment answering a question agencies had been guessing at: which light spectrum actually helps a driver see a pedestrian at night, and how far that finding transfers to other roads.
WISDOT RESEARCH PROJECT 0092-25-23 · UW–MILWAUKEE · UW–MADISON · TAPCO
Blue-spectrum lighting let drivers detect pedestrians ~40 ft sooner
I authored the winning proposal, designed the experimental protocol, built the deep-learning eye-tracking pipeline that measures driver recognition frame by frame, and led data collection across 281 nighttime rounds at two Wisconsin sites — through rain, wet pavement and late-season cold.
An instrumented probe vehicle approached each crosswalk while a mannequin on a remote-controlled skateboard crossed, producing repeatable pedestrian events on live public roads. Drivers were never told when or where a crossing would happen. A CNN pipeline located the exact frame each driver first recognized the pedestrian from infrared face video, then synchronized it against 125 Hz GPS, speed and brake records to recover reaction time and detection distance.
What I built
The measurement stack: synchronized roadway video, driver-facing infrared video, brake logging and RTK-grade positioning, plus the CNN face-landmark and gaze pipeline that turns raw driver video into a defensible recognition timestamp. Every eye onset was verified against three independent records before it entered the analysis.
Why it matters beyond one crosswalk
The result is not a single site recommendation. The study produced a transferable planning framework: scale mounting height and lateral offset with roadway width, predict detection distance from a regression model, then screen that prediction against AASHTO stopping sight distance before anything gets installed.
Spectrum comparison — Milwaukee site, 177 rounds
| Configuration | Reaction time | Detection distance | vs. baseline |
|---|---|---|---|
| Baseline — no enhanced lighting | 0.92 s | 242.0 ft | — |
| 4000 K neutral white · agency reference | 0.72 s | 237.2 ft | −4.8 ft |
| 5000 K cool white | 0.82 s | 239.4 ft | −2.6 ft |
| 6500 K daylight white | 0.75 s | 244.6 ft | +2.6 ft |
| Blue Baseline | 0.73 s | 256.4 ft | +14.4 ft |
| Ice Blue — recommended | 0.69 s | 260.8 ft | +18.8 ft |
| Sky Blue | 1.13 s | 254.8 ft | +12.8 ft |
| Pure Royal Blue | 0.83 s | 239.9 ft | −2.1 ft |
Full protocol, Madison results, eye-tracking physiology and design guidance: iamfahad289.github.io/Enhanced-Light-Crosswalk-Illumination
Research programs
Perception you can put a number on
Detection benchmarks say a model found a box in a frame. None of that tells you whether the resulting trajectory was accurate enough to make a safety decision, or whether the sensor could see anything at all under glare. My work builds the datasets and evaluation frameworks that close that gap.
CrossTraj — pedestrian–vehicle trajectory dataset
A bi-season, multi-camera dataset of pedestrian and vehicle trajectories at a midblock crosswalk in Milwaukee, built with roadside cameras and deep-learning tracking, with RTK GNSS as ground truth. It supports trajectory-level evaluation rather than frame-level detection scores, which is what safety analysis actually requires.
Evaluation framework for roadside perception
A closed-loop framework that scores a roadside perception system against surveyed ground truth end to end — detection through tracking through reconstructed trajectory — so an agency or vendor can state how much positional error their infrastructure sensing carries before it feeds a warning system.
CrossRisk — interaction risk classification
Crash records miss near-misses, so the true risk at a crosswalk is systematically under-counted. CrossRisk classifies each pedestrian–vehicle interaction as safe, risky or critical using gap acceptance, spatial proximity, stopping sight distance, driver behavior and crosswalk encroachment. Validated on 206 real interactions in Milwaukee.
Rural autonomy — USDOT TRAVELS Center
Rural roads carry a fraction of U.S. traffic but 47% of roadway fatalities. Within UW–Milwaukee's $1.4M share of the $15M USDOT TRAVELS Center, I design scaled autonomous vehicle testbeds that reproduce rural failure conditions — unlit stretches, gravel, faded markings, harsh weather — and evaluate how perception and control degrade there.
GLANCE — nighttime glare exposure dataset
Six in ten U.S. drivers call headlight glare a problem after dark, yet glare appears in only 0.1–0.2% of nighttime crash records — because nobody measures the light that actually reaches the driver's eye. GLANCE pairs forward roadway video with an eye-level photometric logger sampling at the spectacle plane at 20 Hz, so a glare event is scored by delivered light rather than by how bright it looks on camera. Calibrating video against the sensor nearly doubled detection of the worst moments (average precision 0.29 → 0.50), and duration rather than peak intensity separated the 58 recorded episodes.
Human-executable connected-vehicle speed advisories
Speed advisory algorithms optimize trajectories for energy and traffic flow, then treat the driver as a disturbance. But drivers never track a prescribed speed exactly — perception limits, reaction delay and longitudinal control error intervene, and the gap widens when an advisory varies continuously or demands fine-grained adjustment. This project treats human executability as a design requirement rather than an error term, optimizing traffic performance, driver tracking limits and advisory simplicity jointly. Field validation planned.
Perception & vision
Estimation & control
Engineering
Scholarly articles
Publications
17 publications · 222 citations · h-index 8 · i10-index 8. Citation counts below are per-paper figures from Google Scholar; also indexed on ORCID 0009-0007-9854-8937.
Under review
Transportation Research Board 2027 Annual Meeting
Field Evaluation of an Enhanced Crosswalk Lighting System Using Pedestrian Detection Distance and Driver Reaction Time
Fahad, M.; Rai, N.; Long, K.; Cao, B.; Ma, C.; et al. · Manuscript submitted to the Transportation Research Board 2027 Annual Meeting
Measuring Nighttime Roadway Glare Using Sensors and Video: Dataset Development, Validation, and Event Analysis
Fahad, M.; Ogunniyi, O. E.; Qin, X. · Manuscript submitted to the Transportation Research Board 2027 Annual Meeting
Geometry-Aware Bounded-Latency Trajectory Extraction for Panoramic Roadside Cameras
Xiong, T.; Fahad, M.; Qin, X.; Zhao, T. · Manuscript submitted to the Transportation Research Board 2027 Annual Meeting
Conference proceedings
Peer-reviewed
Evaluating Roadside Perception with Trajectory Data from Pedestrian–Vehicle Interactions at Crosswalks
Fahad, M.; Tasnim, A.; Xiong, T.; Damaraju, A.; Zhao, T.; et al. · ASCE International Conference on Transportation & Development (ICTD) 2026
A Comprehensive Evaluation Framework for Roadside Perception Systems
Fahad, M.; Tasnim, A.; Xiong, T.; Damaraju, A.; Zhao, T.; Qin, X.; Shi, X. · Transportation Research Board Annual Meeting · National Academies of Sciences, Engineering, and Medicine
Evaluating Crosswalk Safety Through Trajectory Analysis of Pedestrian Gap Acceptance and Vehicle Yielding
Tasnim, A.; Fahad, M.; Xiong, T.; Shi, X. · ASCE International Conference on Transportation & Development (ICTD) 2026
MPILC-based energy management strategy for series–parallel plug-in hybrid electric vehicle
Muhammad Fahad; Liu, C.; Li, L. · 14th International Symposium on Advanced Vehicle Control (AVEC 2018), Beijing, China
Journal articles
Peer-reviewed · 222 citations
Yue, W.; Liu, C.; Li, L.; Chen, X.; Muhammad, F. · Proceedings of the IMechE, Part D: Journal of Automobile Engineering, 235(7), 1894–1904 · Cited 19 times
Liu, C. Z.; Li, L.; Yong, J. W.; Muhammad, F.; Cheng, S.; Wang, X. Y.; Li, W. B. · IEEE Transactions on Intelligent Transportation Systems, 22(12), 7860–7874 · Cited 27 times
Liu, C. Z.; Li, L.; Yong, J. W.; Muhammad, F.; Cheng, S.; Wu, Q. · IEEE Transactions on Intelligent Transportation Systems, 22(11), 7102–7114 · Cited 18 times
Liu, C. Z.; Li, L.; Yong, J. W.; Muhammad, F.; Cheng, S. · IEEE Transactions on Intelligent Transportation Systems, 22(3), 1553–1561 · Cited 13 times
Guo, H. Q.; Liu, C. Z.; Yong, J. W.; Cheng, X. Q.; Muhammad, F. · IEEE Access, 7, 71323–71334 · Cited 43 times
Du, S.; Yang, Y.; Liu, C.; Muhammad, F. · Proceedings of the IMechE, Part D: Journal of Automobile Engineering, 233(7), 1760–1772 · Cited 8 times
Jiang, G.; Miao, X.; Wang, Y.; Chen, J.; Li, D.; Liu, L.; Muhammad, F. · Proceedings of the IMechE, Part D: Journal of Automobile Engineering, 231(14), 1963–1972 · Cited 31 times
Jiang, G.; Liu, L.; Guo, C.; Chen, J.; Muhammad, F.; Miao, X. · Proceedings of the IMechE, Part D: Journal of Automobile Engineering, 231(10), 1380–1390 · Cited 29 times
Li, L.; Wang, X.; Hu, X.; Chen, Z.; Song, J.; Muhammad, F. · ASME Journal of Dynamic Systems, Measurement, and Control, 138(6), 061005 · Cited 30 times
Zhang, L.; Yu, W.; Zhao, X.; Meng, A.; Muhammad, F. · Proceedings of the IMechE, Part D: Journal of Automobile Engineering, 230(11), 1515–1526 · Cited 4 times
Awards & prizes
Recognition
Outstanding Graduate Student Award — 2026
University of Wisconsin–Milwaukee
Institution-level recognition for research at the intersection of artificial intelligence, autonomous vehicle systems, computer vision and traffic safety.
Student of the Year — two consecutive years
Tsinghua University, Beijing
Awarded in back-to-back years at a university ranked among the top 20 worldwide, competing against an international graduate cohort.
Graduate Student Excellence Fellowship
University of Wisconsin–Milwaukee
Competitive fellowship supporting doctoral research in autonomous and safety-critical systems.
3rd Place — Graduate Research Poster Competition
University of Wisconsin–Milwaukee
Placed against graduate research from across the university for work on roadside perception and crosswalk safety evaluation.
China Scholarship Council award
Government of China
Nationally competitive government scholarship supporting graduate study at Tsinghua University.
Selected for presentation at TRB 2026
National Academies of Sciences, Engineering, and Medicine
Two first-author papers accepted to the world's largest gathering of transportation researchers, presented in Washington, D.C. UWM announcement →
Published material about the work
Press & media coverage
Broadcast television segment on the autonomous vehicle research program and its implications for road safety.
National syndication of the UW–Milwaukee autonomous vehicle research coverage.
Extended interview on rural autonomous vehicle research and the design of scaled AV testbeds within the USDOT TRAVELS Center.
Interview covering the crosswalk lighting evaluation project, trajectory-based risk classification, and moving research into public infrastructure decisions.
Autonomous vehicle trade coverage of the project's objectives and methodology.
University reporting naming both first-author papers presented at the Transportation Research Board Annual Meeting.
Syndicated distribution of the TMJ4 broadcast coverage.
Leading & critical roles
Where I carry responsibility
Funded project leadership, the proposals that win that funding, peer review, and elected office at two research universities.
Lead researcher — WisDOT 0092-25-23
Wisconsin Department of Transportation · ~$125K
Authored the winning proposal, then led experimental design, instrumentation, nighttime field deployment across two cities, the eye-tracking analysis pipeline, and the design guidance delivered to the WisDOT Project Oversight Committee. Demonstrated the system live to the committee at both sites and at the SE Wisconsin Transportation Symposium 2025.
Graduate research lead — USDOT TRAVELS Center
$15M USDOT center · $1.4M UW–Milwaukee share
Responsible for the scaled autonomous vehicle testbed and experimental framework at UW–Milwaukee: hardware platform, control software, test scenarios and the data pipeline that analyzes how perception and control behave under rural conditions.
President — Pakistan Student Association
UW–Milwaukee, 2026–present · Tsinghua University, 2016–2020
Elected to lead the association at two research universities, on two continents. Coordinates student programming, cross-institutional events and support for incoming international students.
Peer reviewer
Judging the work of others
Reviews submissions for journals and conferences across intelligent transportation systems, computer vision and autonomous systems, evaluating methodology, experimental validity and contribution.
Professional memberships
IEEE · Transportation Research Board, National Academies of Sciences, Engineering, and Medicine · Institute for Physical Infrastructure and Transportation (IPIT), UW–Milwaukee.
NSF I-Corps — technology commercialization
National Science Foundation · Milwaukee I-Corps program
Completed the full five-workshop NSF I-Corps curriculum, running structured customer discovery to test the commercial pathway for the crosswalk illumination technology. A patent application on the innovation is in preparation.
Research proposals authored
I write the proposals that bring competitive research funding into the lab — scoping the technical approach, experimental design, budget and deliverables. The crosswalk illumination proposal was awarded and delivered; the rest span agency programs, industry solicitations and national innovation competitions.
| Proposal | Program / sponsor | Focus |
|---|---|---|
| Enhanced crosswalk illumination | Wisconsin DOT · 0092-25-23 | Nighttime pedestrian visibility, spectral evaluation |
| Vehicle headlight color and glare | AAA Foundation for Traffic Safety | Headlight spectrum effects on glare and driver safety |
| ADAS evaluation solicitation response | Industry request for quotation | Advanced driver assistance system testing and validation |
| Catalyst technology translation | UWM Research Foundation Catalyst Grant | Commercialization pathway for crosswalk sensing technology |
| Applied AI challenge entry | NVIDIA Challenge | GPU-accelerated perception for autonomous systems |
| Vision-language models for transportation | TRB IDEA · National Academies | Vision-language models applied to roadway scene understanding |
Experience
Track record
- Built CrossTraj, a multi-camera and RTK GNSS benchmark dataset for trajectory-level evaluation of roadside perception systems.
- Designed real-time detection and tracking pipelines (YOLOv9e, DeepSORT, RTK sensor fusion) for safety-critical object tracking and trajectory reconstruction.
- Developed closed-loop evaluation frameworks for pedestrian–vehicle interaction modeling in mixed traffic.
- Contributed to a $1.4M USDOT-funded rural autonomous vehicle deployment initiative.
- Authored peer-reviewed publications in intelligent transportation, trajectory modeling and safety-critical autonomous systems.
Represents Pakistani students and supports international student engagement through cultural, academic and community initiatives across campus.
- Owned system-level architecture for safety-critical automotive platforms.
- Integrated perception, prediction and control pipelines in real-time environments.
- Directed validation strategy across simulation and field testing.
- Defined the technical roadmap for scalable intelligent vehicle systems.
- Deployed real-time machine learning models for trajectory prediction and adaptive vehicle control.
- Integrated AI decision-making modules inside safety-constrained control systems.
- Built large-scale simulation environments for multi-agent interaction testing.
- Implemented performance monitoring and model validation frameworks under operational constraints.
- Led research on AI-driven decision-making for mixed-traffic autonomous environments.
- Designed multi-agent behavior models for vehicle–pedestrian interaction analysis.
- Developed predictive control frameworks combining learning-based and physics-based models.
- Mentored junior researchers and coordinated interdisciplinary projects.
- Designed control algorithms for automated driving under dynamic traffic conditions.
- Built simulation-based validation frameworks for safety and performance evaluation.
- Integrated perception outputs with planning and control modules on experimental platforms.
Led student coordination and cultural programming at one of Asia's leading research universities, building connections between the Pakistani student community and the wider international student body.
Education
PhD, AI for Autonomous and Safety-Critical Systems
University of Wisconsin–Milwaukee · 2024–present
Doctoral research in perception, trajectory modeling and safety validation for autonomous systems. GPA 3.94/4.0.
MS, Mechanical Engineering
Tsinghua University, Beijing
Intelligent vehicle systems, vehicle dynamics, control and energy management for hybrid electric powertrains.
BE, Mechanical Engineering
Sarhad University of Science & IT, Peshawar
Foundations in mechanical systems, control theory and robotics.
Contact
Open to research collaboration and industry roles
I'm particularly interested in work on perception validation, vulnerable-road-user safety, and moving autonomy from benchmark performance to deployed reliability.
Get in touch
For collaboration, dataset access, review requests or hiring conversations, email is the fastest route. Typical response within 48 hours.
Milwaukee, Wisconsin, United States
Profiles
- Google Scholar222 citations · h-index 8
- ORCID0009-0007-9854-8937
- LinkedIniamfahadusa
- GitHubiamfahad289
- ResearchGatePublications
- Crosswalk studyWisDOT 0092-25-23