VLN-CE

Why we included this project

This repo is the reference implementation for Vision-and-Language Navigation in Continuous Environments (VLN-CE), a research task where an agent follows natural-language instructions while moving freely through a photorealistic 3D scene instead of hopping between predefined nodes. That distinction matters a lot for embodied AI and instruction-following work, and it has made VLN-CE a common ground for comparing models. The code bundles the official baseline agents, training methods, and data loaders for both the Room-to-Room (R2R) and Room-Across-Room (RxR) datasets atop the Habitat simulator, and it anchors the recurring VLN-CE and RxR-Habitat benchmark challenges. If you are building navigation agents, this is a practical starting point that saves you from wiring up dataset downloads, episode management, and evaluation from scratch.