GENIE: A Generalizable Navigation System for In-The-Wild Environments
Jiaming Wang, Diwen Liu, Jizhuo Chen, Jiaxuan Da, Nuowen Qian, Minh Man Tram, Harold Soh
AI summary
Problem
Existing navigation systems struggle to generalize across diverse terrains, weather conditions, and sensor noise in unstructured outdoor environments, often failing under real-world domain shifts.
Approach
The system fine-tunes a vision foundation model for pixel-wise traversability prediction and introduces a path fusion strategy that clusters and merges candidate trajectories to improve planning stability in ambiguous settings.
Key results
- Won first place in the Earth Rover Challenge with 79% of the maximum score
- Outperformed the second-best team by 17% across six countries and three continents
- Released a geographically diverse traversability dataset of 15,347 annotated road images
- Achieved state-of-the-art zero-shot traversability prediction on a new multi-continental benchmark
Why it matters
It establishes a new benchmark for robust, generalizable outdoor robot navigation and provides open-source tools to advance real-world embodied AI deployment.
Abstract
Reliable navigation in unstructured, real-world en- vironments remains a significant challenge for embodied agents, especially when operating across diverse terrains, weather con- ditions, and sensor configurations. In this paper, we introduce GeNIE (Generalizable Navigation System for In-the-Wild Envi- ronments), a robust navigation framework designed for global deployment. GeNIE integrates a generalizable traversability pre- diction model with a novel path fusion strategy that enhances planning stability in noisy and ambiguous settings. We deployed GeNIE in the Earth Rover Challenge (ERC) at ICRA 2025, where it was evaluated across six countries spanning three continents. GeNIE took first place and achieved 79% of the maximum possible score, outperforming the second-best team by 17%, and completed the entire competition without a single human intervention. These results set a new benchmark for robust, generalizable outdoor robot navigation. We have released the codebase, pretrained model weights, and the datasets at https://clear-nus.github.io/genie/ to support future research in real-world navigation.