Event

Omni-IGN: Omnidirectional Vision Image-Goal Navigation System for UAVs

1 reports · 1 sources

Zichen Yan and other authors propose Omni-IGN, a fully onboard image-goal navigation (ImageNav) system for UAVs that integrates four-fisheye perception, an attention-based reinforcement learning policy, and a collision-aware safety layer. The authors state that four-fisheye RGB images can simultaneously support ImageNav and obstacle avoidance without the need for dedicated depth sensors. The authors state that flight tests in IsaacSim verified the effectiveness of combining learned depth estimation with an action safety layer for obstacle avoidance; real-world experiments demonstrated zero-shot sim-to-real transfer without fine-tuning or external localization. Code and model checkpoints will be released after the paper is accepted.