TreasureX Researchcs.AILearning Adaptive Safety Margins for Visual Navigation
Junyi Hu, Shuaihang Yuan, Geeta Chandra Raju Bethala, Anthony Tzes, Yi Fang · July 20, 2026
A new paper from NYU and NYU Abu Dhabi tackles a common robot navigation problem: fixed safety margins are either too conservative (causing detours and timeouts) or too permissive (leading to risky shortcuts). The authors propose a context-conditioned safety critic that learns an adaptive clearance preference for ranking trajectory proposals from a diffusion-based planner. The critic decomposes safety into three terms: a safety term with clearance-budget penalty and control-barrier-function residual, an efficiency term with smoothness and safety-gated detour-ratio penalties, and a distance-constraint matching term to prevent margin collapse. The model is trained with privileged ESDF geometry in simulation and distilled into a perception-only selector via teacher-student learning. On PointGoal navigation in HM3D and MP3D, including cross-dataset transfer, it achieves state-of-the-art success rate and SPL. The approach transfers zero-shot to a Unitree G1 humanoid robot navigating cluttered indoor scenes. This matters because it enables robots to navigate more efficiently and safely without manual tuning, reducing development time and improving performance in real-world applications like home assistance or warehouse logistics. The economic implication is lower deployment costs and faster adoption of autonomous navigation in cluttered environments.
Read more: https://arxiv.org/abs/2607.18200
Educational content only, generated with AI assistance from public research abstracts. Not investment advice. Thank you to arXiv for use of its open access interoperability.