Track manipulation, locomotion and embodied systems.
Search papers, labs, and topics across Lattice.
To resolve critical failure modes in continuous round-trip vision-language navigation (VLN)—specifically directional observability, deviation recovery, and termination instability—the authors implement a reliability-aware sparse route memory that queries ordered geometric anchors in reverse. Evaluated on 50 paired episodes using NaVILA and a simulated Unitree Go2 quadruped, the system improves return success from 22.0% to 55.1%, reaching 86.0% under oracle route conditions. The findings reveal that the primary bottleneck in closed-loop return execution is not mere memory retention, but the lack of sufficiently confident geometric evidence needed to trigger online intervention.
Language-guided robots fail round-trip return runs over 75% of the time, but gating reverse execution through reliability-arbitrated geometric anchors more than doubles return success.
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.