Attend a seminar on fine-grained multimodal learning, Sept. 25

About

Venkata Rani is a doctoral student in robotics and autonomous systems in the School of Manufacturing Systems and Networks, part of the Ira A. Fulton Schools of Engineering at Arizona State University. His research focuses on language-driven robot control and fine-grained multimodal understanding for robotics and industrial automation. He develops methods for code generation in domain-specific robot programming languages and studies generative AI models for industrial applications where limited data and strict precision requirements make accuracy and verifiability essential. Rani also gained industry research experience as a software engineering intern at Google, where he developed AI-driven methods for AI infrastructure optimization. His long-term research interests include advancing fine-grained multimodal understanding and exploring how generative AI systems can improve recursively at scale.

Abstract

Automated robotic inspection is expanding across manufacturing, including applications that read analog pressure gauges. These gauges remain common in semiconductor subfabs, petrochemical plants and other industrial environments because of their reliability and low maintenance costs, but they often require manual inspection. Traditional computer vision systems rely on multistage pipelines in which errors can propagate and compromise the final reading. Vision language models offer an alternative by reasoning directly from an image, but their visual understanding often lacks the precision required for industrial measurement.

This talk examines how reward modeling can improve fine-grained vision-language understanding. It highlights reinforcement learning-based post-training, composite rewards that supervise both accuracy and reasoning, and expert rationale distillation. Together, these approaches show how reward-centered post-training can improve compact open-source models for precise, visually grounded industrial inspection.

Fine-grained Multimodal Learning for Visual Instrument Interpretation Using Visual Language Models seminar
Friday, Sept. 25, 2026
10:30 a.m.–noon
Interdisciplinary Science and Technology Building 12 (ITSB 12) room 215, Polytechnic campus [map]