Research
A guided map of the scientific program.
Each research direction begins with a physical problem, states the comparison that could disprove it, and keeps planned experiments separate from results.
CLAIMS REMAIN HYPOTHESESEvidence spine
One architecture. Related research questions.
Recorded physical interaction feeds a common evidence core, then branches into bounded research extensions.
The research directions share recorded interaction data, evidence rules, and a common training architecture; they are not eight unrelated ideas.
Research extensions
Eight questions. Permanent paths into the detail.
Adaptive Sensor Attention
Can a model change how strongly it relies on each available sense from moment to moment?
Open the deep dive ↗02RESEARCH EXTENSIONHierarchical coherence
Does explicit and recursively refined comparison structure add value beyond hierarchy learned implicitly by a strong model?
Open the deep dive ↗03WORKING REPRESENTATION HYPOTHESISMovement vocabulary and packets
Can small pieces of physical interaction become reusable units that transfer across objects and tasks?
Open the deep dive ↗04PROPOSED HYPOTHESISLearning from a few successful attempts
What reusable relationships can be learned from the unavoidable variation among only a few successes?
Open the deep dive ↗05RESEARCH EXTENSIONTwo directions of learning
Can physical interaction produce a policy from which a useful predictive world model can later be recovered?
Open the deep dive ↗06RESEARCH DIRECTIONAgent structure discovery
Can bounded computational agents discover how learning information should be connected, grouped, weighted, stabilized, and extended?
Open the deep dive ↗07RESEARCH EXTENSIONRecovery under irreversible contact
Can a learning system recover useful structure from contact events that permanently change the state of the world, or does it only succeed when the physical trajectory remains reversible?
Open the deep dive ↗Language at every layer
I asked my LLM to clean my room. It built me a Castle. Coming sooner then you think.