Research

My research spans robot learning and multimodal human behavior understanding.

Vision-Language-Action models for robotics

Vision and language inputs driving a VLA model that controls a robot arm

I adapt and evaluate robot foundation models on real manipulation platforms. My focus is on fine-tuning strategies, evaluation metrics that go beyond task success rate, and making policies more reliable at deployment. This is my current work at KDDI Research.

Multimodal affective computing

Video, audio, physiological and text signals fused to predict emotion, engagement and personality

I model human emotion, engagement, and personality from audio, video, text, and physiological signals such as EEG and ECG. This line of work, developed during my Ph.D. at IIT Roorkee, has appeared at ICML, ACM Multimedia, ICPR, FG, and IJCNN, and in Elsevier BSPC.

Vision-Language Models

An image and a question processed by a VLM to produce a description

I work on multimodal understanding with VLMs, which I began during my research internship at Samsung Research Institute Bangalore.

See all papers on the Publications page.