Sparse route memory improves round-trip vision-language robot navigation reliability
Reliability-Aware Sparse Route Memory for Round-Trip Vision-Language Navigation
Robotics
Summary
Robots that navigate by understanding both visuals and language usually only plan one-way trips. The authors study robots that need to go to a place and then come back safely. They designed a method that lets a robot remember key points along its original path and use this memory to better find its way back. Their approach made the robot much more successful at returning, though some challenges remain due to unreliable environment details.
What this means in practice
- •For robotics engineers: Enable robots to better retrace their outbound path and return safely in complex environments using sparse route memories.
- •For warehouse automation teams: Improve automated robots’ ability to complete delivery and then reliably return to charging stations or start points in large facilities.
Authors
Bojun Long, Lingfan Bao, Tianhu Peng, Jingcheng Sun, Chengxu Zhou
Abstract
Vision-language navigation (VLN) is typically evaluated as a one-way task, although deployed robots may need to return after reaching a goal. We study continuous round-trip VLN and diagnose failures in directional observability, deviation recovery, and termination stability. We propose a reliability-aware sparse route memory that records the executed Outbound trajectory as ordered geometric anchors and queries them in reverse through a structured hint, action-level arbitration, and terminal verification. On 50 reverse-paired episodes using NaVILA and a simulated Unitree Go2, language-only Return succeeds in 22.0% of episodes, while our online system reaches 55.1%. With exact route information, the same interfaces achieve 86.0%, showing that effective Return requires both accurate information and consistent action on that information. The remaining online gap arises mainly from geometric evidence that is too unreliable to authorise intervention. These results distinguish information quality, behavioural consistency, and online reliability as separate limits in long-horizon navigation.