A finite-state memory of personal routines guides a VLM's task/NO_ACT choice; a VLA executes. Daily reflection edits the memory, not model weights.
Editorial summary · not the author abstract. Read the linked paper for the complete abstract and claims.PAM overview of task prediction, VLA execution and procedural memory reflection Original figure · paper authors ↗
Release evidence
Linked repository is reachable but contains project-page assets, demos, an example FSM and helper tools—not the complete robot implementation or an evaluation dataset.