Guided Diffusion with Distilled Vision-Language Reliability for Aerial Navigation

2026-06-11

AI Summary

This work addresses the limitations of conventional drone navigation systems, where sequential perception, mapping, and planning stages propagate errors, and end-to-end models often fail to recognize unreliable observations—such as glass surfaces or overexposed regions—leading to hazardous trajectories. The authors propose a reliability-aware diffusion planner that, for the first time, distills the open-vocabulary reasoning capability of vision-language models into real-time reliability heatmaps to guide 3D trajectory generation within a diffusion framework. A differentiable two-stage ESDF-based cost function is introduced to jointly account for physical obstacles and perceptually unreliable regions. Without requiring environment-specific retraining, the method generalizes effectively, reducing obstacle violation rates from 40.3% to 9.6% and improving average trajectory reliability from 0.588 to 0.925 on both simulated and real quadrotor platforms, with the distillation strategy achieving up to a 2× speedup.

Abstract

Autonomous UAV navigation is conventionally solved by pipelines that separate perception, mapping, and planning into distinct stages, which propagates errors, accumulates latency, and requires environment-specific retuning. End-to-end generative models remove these interfaces by mapping raw observations directly to trajectories, but inherit a subtle failure mode: trained on clean data, they cannot recognise when an observation is unreliable, and treat degraded regions such as glass, mirrors, and overexposed surfaces as valid evidence for planning. We present a reliability-aware diffusion planner for 3D UAV navigation. It conditions trajectory generation on the observation together with a scene-level reliability heatmap that marks where perception cannot be trusted, produced by a lightweight network that distils the open-vocabulary reasoning of a vision-language model within the real-time planning budget. To generalise to unseen environments without retraining, we steer the denoising process with a differentiable two-stage ESDF cost that treats physical obstacles from depth and virtual obstacles from highly unreliable regions on equal footing. In simulation and on a real quadrotor, our planner produces markedly safer trajectories than a state-of-the-art diffusion baseline, reducing the obstacle-violation rate from 40.3% to 9.6% and raising the mean reliability of traversed regions from 0.588 to 0.925. Ablating the reliability term alone drops mean reliability from 0.898 to 0.783, confirming it as the decisive component, while distillation runs the framework up to 2 times faster than the full vision-language model.

Problem

• autonomous UAV navigation
• unreliable observations
• end-to-end generative models
• perception reliability
• trajectory planning

Innovation

• reliability-aware diffusion
• vision-language distillation
• aerial navigation
• ESDF cost
• end-to-end trajectory planning