Target-Aligned Fusion for Decision-Sequence Learning under Dynamics Shift

Understand

External trajectories can improve offline decision-sequence learning, but dynamics shift may make some source subsequences inconsistent with the target environment.

• We study how to fuse such trajectories with limited target data for Decision Transformer learning under dynamics shift.
• We propose Target-Aligned Fusion (TAF), a principled framework that derives source-data fusion from a target-domain Bellman-risk criterion.
• Our analysis bounds this risk by two measurable data-alignment quantities: $\Delta_m$, the state-structure mismatch of retained fragments, and $\Delta_w$, the weighted transport cost from source to target transitions.