StateLinFormer Uses Stateful Training for Long-term Navigation Memory

StateLinFormer employs a stateful memory mechanism that preserves recurrent memory states across training segments rather than reinitializing at batch boundaries. This linear-attention approach allows navigation models to approximate learning on infinite sequences, outperforming standard Transformers in long-horizon tasks.