Vision language model guides robot navigation by comparing options first
C$^2$Nav: Compare Before You Commit for Zero-Shot Vision-and-Language Navigation
Robotics
Summary
Robots that follow instructions to move around can struggle when asked to decide exactly where to go next. The authors propose a different way: instead of asking the robot to pick a direction outright, they let it compare several possible paths before choosing. This method keeps the robot’s movement flexible and reversible, improving how well it follows instructions without needing extra training. Their tests show this comparison approach helps the robot stop in the right places and follow routes more reliably.
What this means in practice
- •For robotics engineers: Build navigation systems where a vision-language model compares possible directions to improve real-time robot path choices without new training.
- •For autonomous vehicle developers: Implement decision interfaces in self-driving systems that use model comparisons over fixed choices to enable safer, reversible turns and stops.
Authors
Runtian Zheng, Congpeng Zhang, Ying Liu
Abstract
Zero-shot vision-and-language navigation in continuous environments (VLN-CE) increasingly places foundation vision-language models (VLMs) inside the navigation loop. Existing systems commonly request cardinal outputs such as waypoints, pixels, headings, progress values, or absolute arrival decisions, coupling a generative response to geometric magnitude or an irreversible commitment. We study a complementary model-robot interface: the VLM compares controller-constructed alternatives, while geometry, thresholds, action magnitude, and execution remain on the physical side. We instantiate this idea in C2Nav, a training-free framework with three coordinated faculties. Seeing performs ordinal Gaze Election over physically vetted candidate views; Remembering maintains a compact route sketch and compares adjacent instruction-leg hypotheses; and Arriving combines a hesitation ladder, look-back comparison, and revocable walk-back for reliable stopping. On the public OpenNav R2R-CE 100 protocol, C2Nav with Qwen3-VL-8B-Instruct obtains 41.0% OSR, 31.0% SR, and 16.7% SPL, while the same interface with the standard GPT-5.5 model reaches 54.0% OSR, 44.0% SR, and 29.0% SPL. Whole-faculty ablations reduce SR to 14.0% without Seeing, 25.0% without Remembering, and 29.0% without Arriving. Matched role inversions that replace only the comparative answer form with cardinal/absolute questions reduce SR to 12.0%, 28.0%, and 21.0% in the spatial, transition, and terminal slots, respectively. The results indicate that a constrained decision interface and stronger VLM reasoning are complementary rather than interchangeable.