Research
I study how intelligent systems can learn from human behavior and neural activity while remaining understandable to the people who supervise them.
Physical AI
My Physical AI work focuses on embodied agents in realistic, high-load settings. I build closed-loop systems in which human feedback—especially implicit error signals—can adapt an agent’s visual, auditory, and tactile guidance. The aim is safer autonomy that keeps people meaningfully involved.
Brain–Computer Interfaces
I develop EEG decoding methods and immersive BCI systems for motor imagery, tactile perception, emotion recognition, and neural–language modeling. A recurring goal is robustness across users: learning representations that transfer while preserving meaningful subject-specific structure.
Explainable AI
I investigate counterfactual explanations for fine-grained recognition and neural decoding. These methods move beyond generic saliency by identifying object- and part-level evidence associated with misclassification, helping researchers inspect failure modes rather than simply visualize attention.
Current direction
My current research brings these strands together: neuroadaptive interfaces for human–embodied AI collaboration. I am interested in international collaborations involving embodied agents, multimodal neural signals, adaptive oversight, and trustworthy machine learning.