I work on real-time BEV perception, radar-vision fusion, and RAW-to-task co-design for autonomous systems, with an emphasis on deployment efficiency, robustness, and sensing-aware learning.
My work connects perception models, sensing design, and deployment constraints. I am interested in systems that can reason reliably under noise, limited compute, and the mismatch between clean benchmarks and field conditions.
I study how to make autonomous perception pipelines smaller, faster, and more robust without giving up task performance. That includes distillation for BEV systems, multimodal fusion, unsupervised restoration, and task-aware sensing pipelines that treat the imaging stack as part of the learning problem.
The goal is practical autonomy: models that work under latency budgets, degraded sensing, and real deployment constraints.
A compact running timeline of papers, awards, and academic progress.
Each thread is tied to deployable autonomy rather than benchmark-only performance.
Distilled and modular BEV pipelines for real-time perception and planning under strict compute budgets.
Unsupervised denoising, restoration, and risk-aware learning when labels are scarce and measurements are corrupted.
Joint optics-sensor-model design that optimizes sensing itself for downstream autonomous tasks.
Reeshad Khan, John Gauch · CVPR, 2026
