Gandhimathi (Mathi) Padmanaban
Ph.D. Candidate - U of M-Dearborn | Limit-Aware Hybrid AI for Transportation Safety
Research Vision: Building hybrid perception systems that know when predictions break physics, and know when to abstain, not just when to be confident.
About Me
I am a Ph.D. Candidate in Industrial and Systems Engineering at the University of Michigan-Dearborn. My dissertation, "Enhancing Transportation Safety: Research on Driver Behaviors Using Advanced Machine Learning," integrates data-driven machine learning, physics-informed methods, and geometry-informed computer vision for transportation safety applications.
My work with both data-driven and physics-informed approaches has revealed limitations in each method individually. This informs my future research on Limit-Aware Hybrid AI frameworks organized around three core thrusts: physically grounded architectures that embed domain constraints as structural components, geometry-informed representation learning that leverages mathematical models of space and motion, and operational awareness with structured abstention that enables systems to know when to defer rather than confidently predict. These frameworks create perception systems that are both statistically accurate and operationally robust for safety-critical deployment. I am committed to open science and reproducible research. Currently seeking academic research positions.
Research Focus
My current research develops geometry-informed computer vision and machine learning methods for transportation safety and driver behavior analysis. My future research vision focuses on Limit-Aware Hybrid AI for reliable deep perception, developing frameworks that embed physical and regulatory limits as structural components of perception architectures to create systems that are both statistically accurate and operationally robust.
Education
Certifications
Research Interests
- Methodological Focus: Limit-aware spatial intelligence • Geometry-informed computer vision • Hybrid data-driven and physics-informed machine learning • Physically grounded architectures • Geometry-informed representation learning • Operational awareness and structured abstention
- Application Domains: Transportation safety and cyclist-vehicle interactions • Driver behavior analysis • Reliable deep perception for autonomous systems • Safety-critical decision support systems