DART-VLN: Test-Time Memory Decay and Anti-Loop Regularization for Discrete Vision-Language Navigation
This episode discusses the DART-VLN paper, which introduces test-time adjustments for Vision-Language Navigation agents. By implementing memory decay through metadata reweighting and adding anti-loop regularization to prevent backtracking, researchers improved navigation success rates and significantly reduced runtime on the REVERIE dataset without needing to retrain existing models.