Blog
Short perspectives on robot learning, drawn from my own work and the papers I keep returning to. I write these to think out loud about where embodied AI is heading.
Jun 12, 2026 · 6 min read
Why retrieval belongs in the robot learning stack
The space of robot tasks is unbounded, but good neighbors are not. A case for retrieving past experience and adapting it, rather than collecting new demonstrations for every task.
May 28, 2026 · 7 min read
World models, and why inverse dynamics beats joint prediction
If the goal is control, recovering the action that explains a transition is often a better use of video than predicting every future pixel.
Apr 30, 2026 · 5 min read
Scene graphs as the interface between perception and action
Most embodied failures are perception failures. Structured, object-centric representations give agents the spatial grounding that flat features miss.