Mature enough for production?

Thanks for sharing your great work!

I’ve seen the impressive SLAM demos from Depth Anything 3. But is it mature enough for real production use in home or warehouse environments? As far as I know, no fully learning-based SLAM system has been deployed in production yet. While learning-based methods can replace individual components—such as feature extraction, depth estimation, or loop-closure detection—the complete end-to-end SLAM pipeline still relies heavily on geometric, model-based algorithms for robustness, interpretability, and long-term stability. In practice, learning-based modules are typically used to augment traditional SLAM rather than replace it entirely.

Please let me know what you think.

Best,
-Deliang