Artificial Intelligence

ViCoS Lab Logo ViCoS Lab
Thesis work 2026

Token Grounding in Vision-Language-Action Models

Language-conditioned queries score the native X-VLA image-token grid to expose which regions matter for the current manipulation step.

ViCoS Lab Logo ViCoS Lab
Thesis work 2026

Expert Distillation from Frozen VLA Representations for Perceptual Guidance

Recover expert perceptual structure from X-VLA representations such as DINOv2-like tokens or CeDiRNet-aligned maps to support action guidance and representation alignment.

PIU identity unlearning examples
Computer Vision Laboratory Logo Computer Vision Lab
Preprint 2026

PIU: Proximity-guided Identity Unlearning in ID-Conditioned Diffusion Models

Targeted identity unlearning for ID-conditioned diffusion models, with emphasis on forgetting requested identity information while preserving generation quality.

Robotics