DART-VLN: Test-Time Memory Decay and Anti-Loop Regularization for Discrete Vision-Language Navigation
arXiv:2607.01043v1 Announce Type: cross Abstract: Memory-based discrete vision-language navigation (VLN) agents must act under partial observability, yet even strong frozen backbones remain vulnerable at test time. Two common failure modes are stale historical evidence at memory readout and inefficient local backtracking during action selection. We present DART-VLN, a training-free test-time control framework for discrete VLN. DART-VLN combines Test-Time Memory Decay, a read-side memory reweighting rule that suppresses stale and redundant evidence without rewriting stored content, with Anti-Lo
The continuous development in AI for navigation and autonomous action necessitates ongoing improvements in handling imperfect information and preventing common failure modes.
Improving test-time robustness and efficiency for vision-language navigation agents is critical for deploying more reliable and capable AI agents in complex, real-world environments.
This framework offers a training-free method to enhance the performance of discrete vision-language navigation agents by addressing issues of stale memory and inefficient local backtracking, making them more resilient.
- · AI agents developers
- · Robotics companies
- · Autonomous system integrators
- · Logistics and delivery platforms
- · Companies relying on less robust navigation AI
- · Developers struggling with test-time model instability
More robust and efficient autonomous navigation for robots and AI systems becomes achievable without expensive retraining.
This could accelerate the deployment of AI agents in dynamic environments, leading to higher adoption in industries like manufacturing and exploration.
Increased reliability of AI agents may foster greater public trust and reduce regulatory hurdles for autonomous system expansion.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI