Skip to main content

ThinkVLN

A pinned visual-language-navigation project exploring instruction following, subtask decomposition, and context-aware embodied agents.

Claude Monet, Cliff Walk at Pourville

ThinkVLN is the main public trace of my earlier embodied-agent work: how navigation agents should reason when instructions, observations, local memory, and changing world state do not line up cleanly.

The project now sits below the multi-agent systems work on this site, but it still shapes how I think about state, memory, and evaluation.