MOST: Memory of Object States and Task Progress for Long-Horizon Mobile Manipulation
MOST is an instruction-conditioned, end-to-end policy for long-horizon mobile manipulation that uses policy-internal memory structured around task progress and object-state changes. It compresses multi-view history into temporal tokens, supervises historical and predictive future tokens with subtask and object states during training, and conditions action generation without an external planner or retrieval stage. The authors introduce RoboCasa-Long and report consistent gains over XR-1 and a visual-history-only variant.
