Postdoctoral Scholar · UCF

Reeshad Khan, Ph.D.

Postdoctoral Scholar · University of Central Florida

Efficient autonomous vehicle perception: sensor co-design, BEV planning, and AI-powered urban mobility.

I design deployment-ready AV perception. TinyBEV achieves 39.0 mAP at 11 FPS (5x faster, 78% fewer parameters), and my RAW-to-task co-design framework gains +6.8 mIoU by jointly optimizing optics, sensor, and segmentation. Postdoctoral Scholar at UCF CECS under Dr. Xishun Liao, now scaling to AI-powered mobility and urban digital twins.

Real-Time BEV Sensor Co-Design Knowledge Distillation Digital Twins Mobility AI VLA Policy
About

Building perception for autonomous vehicles that works in the real world.

My work spans the full AV perception stack, from jointly optimizing optics and neural models to distilling large teachers into edge-deployable BEV networks that maintain safety metrics under sensor degradation.

I build autonomous vehicle perception systems that are small, fast, and field-deployable. My Ph.D. work produced TinyBEV, a camera-only BEV model distilled from UniAD reaching 39.0 mAP and 0.32 collision rate at 11 FPS, 5× faster with 78% fewer parameters than the teacher. A parallel RAW-to-task co-design framework jointly optimizes optics, color filter array, and a ~1M-parameter segmentation head, gaining +6.8 mIoU over fixed pipelines at ~28 FPS on KITTI-360.

As a Postdoctoral Scholar at UCF CECS under Dr. Xishun Liao, I am now extending this work to AI-powered urban mobility: spatiotemporal trajectory mining, generative human/vehicle digital-twin simulation, and controllable scenario generation for autonomous vehicle testing at city scale.

AV Perception BEV distillation · RAW-to-task co-design · VLA policies
Key Results TinyBEV: 39.0 mAP · 11 FPS · 5× faster  |  L2S: +6.8 mIoU
Current Work · UCF AI-powered mobility · spatiotemporal modeling · urban digital twins
Education

Academic background

2021 – 2026
Ph.D., Computer Science University of Arkansas, Fayetteville, AR Dissertation: Efficient Deep Neural Networks for Autonomous Perception Advisor: Dr. John M. Gauch  ·  Defended December 8, 2025
2024
M.S., Computer Science University of Arkansas, Fayetteville, AR
2021
M.S., Computer Science and Technology Chang'an University, Xi'an, China
2012
B.S., Computer Science and Engineering Bangladesh
Recent Activity

News, acceptances, and milestones.

A compact running timeline of papers, awards, and academic progress.

Jul 2026
🏢
Joining the University of Central Florida (CECS) as a Postdoctoral Scholar under Dr. Xishun Liao, focusing on AI-powered mobility, digital twins, and spatiotemporal trajectory modeling.
Jun 2026
🚗
New preprint: Beyond Bayer - Task-Optimal Sensor Co-Design for Robust Autonomous Driving Segmentation.
Mar 2026
🏆
TinyBEV accepted as a poster at CVPR DriveX Workshop 2026 (avg. reviewer score 8.0/10).
Jan 2026
📄
L2S Driving arXiv preprint released and submitted to ECCV Workshop 2026: Learning to Sense for Driving.
Jan 2026
✨
Paper published: Adaptive Extensions of Unbiased Risk Estimators for Unsupervised MRI Denoising (CVC 2026).
Dec 2025
🎓
Successfully defended Ph.D. dissertation: Efficient Deep Neural Networks for Autonomous Perception.
Jan 2025
🚗
Paper published: TinyBEV (ICCV WDFM 2025).
Jan 2025
🧠
Paper published: From Noise Estimation to Restoration (VISAPP 2025).
Jan 2025
🧾
Paper published: Learning From Oversampling (IEEE Access 2024).
Jan 2024
🏅
Award: Reginald R. Baxter Graduate Fellowship (2024).
Jan 2023
🏅
Award: EECS Graduate Fellowship, University of Arkansas (2023).
Jan 2022
🏅
Award: College of Engineering Graduate Fellowship, University of Arkansas (2022).
Jan 2019
🌏
Scholarship: Chinese Government Belt and Road Scholarship (2019).
Research

Four problem spaces shaping the work.

Each thread is tied to deployable autonomy, from sensor to city, not just benchmark performance.

01

Efficient 3D perception and BEV autonomy

Distilled and modular BEV pipelines for real-time perception, planning, and VLA policies under strict compute budgets and sensor degradation.

02

Robust learning under noisy sensing

Unsupervised denoising, restoration, and risk-aware learning when labels are scarce and measurements are corrupted.

03

Sensing-aware co-design

Joint optics-sensor-model design that optimizes the imaging stack (CFA, noise model, and segmentation head) for downstream autonomous tasks.

04

AI-powered mobility and digital twins

Spatiotemporal trajectory mining, generative AI for human/vehicle digital-twin simulation, and controllable urban scenario generation.

Selected Work

Publications

Featured publication

Learning to Sense for Driving: Joint Optics-Sensor-Model Co-Design for Semantic Segmentation

Reeshad Khan, John M. Gauch · ECCV Workshop, 2026 (arXiv preprint)

RAW-to-task Sensing Segmentation
Read Paper Under review
Preview for Learning to Sense for Driving: Joint Optics-Sensor-Model Co-Design for Semantic Segmentation
6 total 4 published 0 accepted 2 in review
Professional Service

Reviewer activity & certifications

Peer Review ICML SCALE Workshop, 2026
Peer Review IEEE ICASSP, 2026
Peer Review ECML-PKDD, 2026
Peer Review IEEE Big Data, 2022
Certification NVIDIA Accelerated Computing with CUDA Python (2025)
Certification Generative AI with Diffusion Models, NVIDIA (2025)
Project Demos

Selected videos

Demo 1
Demo 2
Mentored Students

Student work and posters I mentored

01Deep Learning Applications in Cyber Security and Cyber Security Demo BoardsKaylynn Charles 02Predicting IDS and IPS attacks using Machine LearningBishop Butler 03Developing Ignition HMI and Enhancing Technological Education with NETLAB+Joseph Dumond 04Canary Historian and Machine LearningCameron Eddy 05Integrating OpenPLC into Cybersecurity Testbed and Revamping Digital Design CourseDaniel Eyraud 06PCB Design in Cadence and Hardware-in-the-Loop SystemsWilliam Graef, Gideon Sutterfield, Anna Corbitt, Wesley Schwartz 07Developing a virtual Enterprise/IT/OT Cyber Test Bed (GHOST)Zachary Harris, Gideon Sutterfield, Henry Schmidt 08Malcolm IDS for Virtualized OT NetworkRyan Thomas Holmes 09Communication Protocols against Attack VectorsIsabella Martinez 10DNP3 Attack Vectors in Arkansas Nuclear One TestbedDavid Nystrom 11GHOST Network: Intrusion Detection System Using Security OnionRichard Stokes 12Side Channel Attack (SCA) of AES in FPGAs and MicrocontrollersDonna Thakadipuram
Visitors

Visitor map and country counter

World Map
Country Counter
Flag Counter
Contact

Open to research collaborations and technical conversations.

re141872@ucf.edu LinkedIn Google Scholar UCF GitHub