# CATNAV: Cached Vision-Language Traversability for Efficient Zero-Shot Robot Navigation

Source: [CATNAV: Cached Vision-Language Traversability for Efficient Zero-Shot Robot Navigation](<https://www.sota2.com/research/paper/69c36c070636e5bf4afc64d5>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-26T05:00:04.122Z

Authors: Aditya Potnis, Francisco Affonso, Shreya Gummadi, Naveen Kumar Uppalapati, Girish Chowdhary

Published: 2026-03-24T04:47:33.000Z

## Abstract

Navigating unstructured environments requires assessing traversal risk relative to a robot's physical capabilities, a challenge that varies across embodiments. We present CATNAV, a cost-aware traversability navigation framework that leverages multimodal LLMs for zero-shot, embodiment-aware costmap generation without task-specific training. We introduce a visuosemantic caching mechanism that detects scene novelty and reuses prior risk assessments for semantically similar frames, reducing online VLM queries by 85.7%. Furthermore, we introduce a VLM-based trajectory selection module that evaluates proposals through visual reasoning to choose the safest path given behavioral constraints. We evaluate CATNAV on a quadruped robot across indoor and outdoor unstructured environments, comparing against state-of-the-art vision-language-action baselines. Across five navigation tasks, CATNAV achieves 10 percentage point higher average goal-reaching rate and 33% fewer behavioral constraint violations.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Robot Navigation on Task 1: Navigate to cone, stay on right side of path (Static)](<https://www.sota2.com/research/sota/robot-navigation-on-task-1-navigate-to-cone-stay-on-right-side-of-path-static.md>): Robot navigation; Task 1: Navigate to cone, stay on right side of path (Static); Goal Reaching: 70.

- [Robot Navigation on Navigate to cone, stay in center of path Task 2 Static](<https://www.sota2.com/research/sota/robot-navigation-on-navigate-to-cone-stay-in-center-of-path-task-2-static.md>): Robot navigation; Navigate to cone, stay in center of path Task 2 Static; Goal Reaching: 70.

- [Robot Navigation on Task 3 Navigate to cone Human crosses path Dynamic](<https://www.sota2.com/research/sota/robot-navigation-on-task-3-navigate-to-cone-human-crosses-path-dynamic.md>): Robot navigation; Task 3 Navigate to cone Human crosses path Dynamic; Goal Reaching Success: 70.

- [Robot Navigation on Navigate to house Task 4 Avoid Collisions with Benches Static](<https://www.sota2.com/research/sota/robot-navigation-on-navigate-to-house-task-4-avoid-collisions-with-benches-static.md>): Robot navigation; Navigate to house Task 4 Avoid Collisions with Benches Static; Goal Reaching: 40.

- [Robot Navigation on Task 5: Go to door, avoid walking over the paper on the floor Static](<https://www.sota2.com/research/sota/robot-navigation-on-task-5-go-to-door-avoid-walking-over-the-paper-on-the-floor-static.md>): Robot navigation; Task 5: Go to door, avoid walking over the paper on the floor Static; Goal Reaching: 90.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2603.22800>)

[PDF](<https://arxiv.org/pdf/2603.22800.pdf>)
