[Under Review] Hindsight semantic exploration for vision-language navigation
Generative hindsight reasoning provides semantic supervisions for on-policy explorations
Sung June Kim
About me
I am a PhD candidate in Artificial Intelligence at Korea University, advised by Prof. Sangpil Kim. I am also fortunate to be advised by Prof. Honglak Lee during my visiting research at the University of Michigan. I enjoy studying how embodied agents interpret human language, perceive their environments, and take actions. Specifically, my primary research focuses on developing data-efficient training methods for language-guided embodied decision-making, with the goal of reducing reliance on human-annotated data.
Selected publications
(Please visit Google Scholar for full publication list with citations.)
[NeurIPS 2025] Active test-time vision-language navigation
Uncertainty-based self-active test-time feedback system for navigation agents
[ICML 2025] Test-time adaptation for online vision-language navigation with feedback-based reinforcement learning
Binary feedback-based RL enables test-time adpatation for navigation agents.
[CVPR 2025] 3D occupancy prediction with low-resolution queries via prototype-aware view-transformation
Mapping of 2D cluster features onto 3D voxel complements the loss of spatial storage for low-resolution queries.
[CVPR 2024] Higher-order relational reasoning for pedestrian trajectory prediction
The rippling indirect social relations that were always there to affect your trajectory