CPNav: Calibrated Vision-Language Navigation for Object-Goal Search in Unseen Environments
CPNav: Calibrated Vision-Language Navigation for Object-Goal Search in Unseen Environments
This paper presents CPNav, a navigation framework combining collision-aware planning with statistically calibrated Vision-Language Model action scoring for map-free object-goal navigation. Using RGB-D input, the system constructs navigability maps and queries two VLMs for goal detection and action scoring, with offline calibration via conformal prediction providing finite-sample guarantees. Evaluated in AI Habitat simulation and on a modified Hiwonder MentorPi robot, CPNav achieves significant relative improvements in success rate and success weighted path length compared to uncalibrated baselines.
The study introduces CPNav, a navigation framework that applies conformal prediction to calibrate Vision-Language Model action scores for object-goal search in unseen environments, demonstrating relative improvements of 6.2% and 11.4% in success rate and success weighted path length respectively on HM3D v0.2 within AI Habitat simulation compared to an uncalibrated baseline.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- code: Available Artifact link Open source S5
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
er-specified risk level. The CP layer is model-agnostic and can wrap any VLM policy. Online, calibrated VLM scores guide action selection, while a tree-structured exploration history records alternative actions and allows backtracking. We evaluate CPNav in both simulation and real-world hardware settings. In AI Habitat simulation, compared with the uncalibrated baseline method, CPNav achieves relative SR and SPL impr
Open source S4
ovements of 6.2% and 11.4% on HM3D v0.2, and 12.4% and 8.8% on MP3D, respectively. In real-world deployment on a modified Hiwonder MentorPi robot, the corresponding improvements in SR and SPL are 24.9% and 15.2%, respectively. Code is available at https://github.com/practice-lab-ucla/CPnav.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org