A reward-free world model that learns spatially structured latent representations reflecting the physical environment, enabling interpretable dynamics and efficient goal-directed planning.

A multi-task learning framework that assigns competing objectives to mutually orthogonal latent subspaces, preventing representation collapse while enabling interpretable and compositional concept manipulation.

@InProceedings{Huang2026DE, 
	author = {Chi-Yao Huang and Khoa Vo and Aayush Verma and Duo Lu and Yezhou Yang}, 
	title = {Domain Expansion: A Latent Space Construction Framework for Multi-Task Learning}, 
	booktitle = {Proceedings of the International Conference on Learning Representations (ICLR)}, 
	year = {2026}, 
}

A visual odometry framework that casts camera-state estimation as contrastive label ranking, learning spatially coherent and interpretable latent representations compatible with multimodal data.

@InProceedings{Huang2026VOCAL, 
	author = {Chi-Yao Huang and Zeel Bhatt and Yezhou Yang}, 
	title = {VOCAL: Visual Odometry via ContrAstive Learning}, 
	booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, 
	year = {2026}, 
}

The Toyota Mothership project aims to develop a versatile, general-purpose aerial robot designed to assist with transportation, communication networks, and disaster prevention.

• Engineered a robust latent-centric pose estimation model for high-altitude aerial robots, leveraging IR and IMU sensor data to ensure reliability across diverse weather conditions.

Simultaneous localization and mapping with zero-shot 3D object reconstruction.

• Zero-shot 3D shape reconstruction from a single image.
• Object recognition, tracking, and mapping.
• Designed a new object-based error function for bundle adjustment.

A* path planning for a biped robot.

• Zero-moment-point motion planning for a biped robot.
• A*-based obstacle avoidance.

VIVE COSMOS is the prototype of all VR, AR, and XR products in VIVE.

• Developed the visual–inertial SLAM system for VR tracking.
• Designed multiple-camera motion estimation and sensor fusion algorithms.

Enterprise-grade mixed reality headset with high-resolution displays and ergonomic comfort.

• Developed the visual–inertial SLAM system for inside-out tracking.
• Designed multiple-camera motion estimation and sensor fusion algorithms.
• Collaborated with UX teams to optimize tracking for dynamic environments.

Lightweight, modular XR headset built for immersive mixed reality experiences.

• Implemented visual–inertial odometry algorithms optimized for low-latency XR.
• Designed and prototyped the core data structures and processing pipeline for a mixed-reality system.
• Engineered an obstacle avoidance system to enhance user safety in mixed-reality environments.

Compact, glasses-style VR headset designed for wellness, streaming, and portability.

• Designed the inside-out tracking algorithm for limited-FoV sensors.
• Developed a prototype motion tracking pipeline integrating a real-time hand-tracking system.
• Collaborated with optics and gaming teams to tune and optimize on-device performance.