StateLinFormer: Stateful Training Enhancing Long-term Memory in Navigation

TL;DR

StateLinFormer introduces a stateful linear-attention model for navigation that maintains long-term memory across interactions, significantly improving long-horizon memory retention and adaptation in navigation tasks.

Contribution

It presents a novel stateful training paradigm for linear-attention models, enabling persistent memory across segments, which enhances long-term navigation capabilities.

Findings

Outperforms stateless and fixed-window Transformers in navigation tasks.

Significantly improves long-horizon memory retention with increased interaction length.

Enhances in-context learning capabilities for navigation.

Abstract

Effective navigation intelligence relies on long-term memory to support both immediate generalization and sustained adaptation. However, existing approaches face a dilemma: modular systems rely on explicit mapping but lack flexibility, while Transformer-based end-to-end models are constrained by fixed context windows, limiting persistent memory across extended interactions. We introduce StateLinFormer, a linear-attention navigation model trained with a stateful memory mechanism that preserves recurrent memory states across consecutive training segments instead of reinitializing them at each batch boundary. This training paradigm effectively approximates learning on infinitely long sequences, enabling the model to achieve long-horizon memory retention. Experiments across both MAZE and ProcTHOR environments demonstrate that StateLinFormer significantly outperforms its stateless linear-attention counterpart and standard Transformer baselines with fixed context windows. Notably, as interaction length increases, persistent stateful training substantially improves context-dependent adaptation, suggesting an enhancement in the model's In-Context Learning (ICL) capabilities for navigation tasks.