My research explores how haptic feedback can facilitate the acquisition and transfer of dexterous sensorimotor skills. Using VR-based training environments and force-feedback devices, I investigate how different feedback strategies influence motor performance, learning, and skill retention.

In particular, I develop AI-driven adaptive haptic assistance that adjusts feedback based on learners' performance and skill levels, aiming to balance direct guidance with opportunities for independent motor learning.

I develop AI-powered multimodal accessibility systems that enable blind and low-vision users to explore and interact with visually represented information, including charts, diagrams, and block-based programming environments.

My work combines AI-based content understanding, refreshable tactile displays, braille, and audio feedback to provide interactive, nonvisual representations. Beyond converting visual content into tactile graphics, I investigate how users can independently navigate, interpret, and manipulate complex information through multimodal interaction.

Presentation slides convey information through spatial arrangements, grouping, and visual hierarchy, which are often difficult for blind and low-vision users to access through conventional screen readers.

In collaboration with Microsoft and Dot Inc., we developed DotVista, an on-device AI-powered accessibility application that transforms presentation slides into interactive tactile representations. Leveraging Microsoft's Windows ML and Windows AI APIs, DotVista analyzes slide content and spatial layouts, providing tactile graphics, braille, and audio descriptions through Dot's refreshable tactile display.

DotVista was showcased at Microsoft Build 2025 and is available on the Microsoft Store. Building on this work, I am exploring shared tactile workspaces to support spatial referencing and collaboration between blind and sighted users.

I explore how AI can enhance audiovisual experiences by generating multisensory effects tailored to individual preferences. Traditional 4D experiences typically rely on predefined effects, offering limited flexibility in adapting to different viewers and their sensory preferences.

To address this limitation, we developed TaME4D, an AI-based system that analyzes audiovisual content and generates personalized 4D effects, including vibration, motion, airflow, and thermal feedback. By combining multimodal scene understanding with natural language descriptions of user preferences, the system automatically identifies relevant scenes and adapts sensory effects to create more immersive and personalized viewing experiences.