NavOL: Navigation Policy with Online Imitation Learning

Code is coming soon. This repository will host the official implementation of NavOL, an online imitation learning framework for visual navigation built on IsaacLab and a pre-trained navigation diffusion policy.

Project Page

🌐 https://logosroboticsgroup.github.io/NavOL/

Paper

NavOL: Navigation Policy with Online Imitation Learning
Xiaofei Wei*, Chun Gu*, Li Zhang
School of Data Science, Fudan University · Shanghai Innovation Institute
ICML 2026 (Accepted)

TL;DR

NavOL fine-tunes a pre-trained navigation diffusion policy (NavDP) via massively parallel rollouts in IsaacLab, supervised online by a privileged global path planner. The rollout–update loop trains on the policy's own visited state distribution, removing the need for reward design and mitigating distribution shift in offline imitation learning. With 8 RTX 4090 GPUs, the system collects more than 2,000 high-quality trajectories per hour across 50 indoor 3D scenes.

What's coming

• Training code (rollout–update loop on top of IsaacLab + NavDP)
• Indoor navigation benchmark on 3D-Front
• Pre-trained checkpoints
• Real-world deployment scripts

Citation

@inproceedings{wei2026navol,
  title     = {{NavOL}: Navigation Policy with Online Imitation Learning},
  author    = {Wei, Xiaofei and Gu, Chun and Zhang, Li},
  booktitle = {Proceedings of the International Conference on Machine Learning (ICML)},
  year      = {2026}
}