# DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras

**Authors:** Zachary Teed, Jia Deng  
**arXiv:** [2108.10869](https://arxiv.org/abs/2108.10869)  
**Date:** 2021-08-24  
**Summary page:** https://www.summarizepaper.com/en/paper/2108.10869v1/

## Key points

- DROID-SLAM is a deep learning-based Simultaneous Localization and Mapping (SLAM) system
- Developed by Zachary Teed and Jia Deng
- Utilizes recurrent iterative updates of camera pose and pixelwise depth through a Dense Bundle Adjustment layer
- Offers highly accurate and robust performance with fewer catastrophic failures compared to previous work
- Can leverage stereo or RGB-D video during test time, even though trained on monocular video
- Open-source code available on GitHub at https://github.com/princeton-vl/DROID-SLAM
- Represents a major advancement in the field of visual SLAM systems
- Enables exceptional accuracy and robustness through its deep learning-based approach and recurrent iterative updates with Dense Bundle Adjustment layer
- Holds great promise for applications requiring precise localization and mapping capabilities

## Summary

DROID-SLAM is a revolutionary deep learning-based Simultaneous Localization and Mapping (SLAM) system developed by Zachary Teed and Jia Deng. This system utilizes recurrent iterative updates of camera pose and pixelwise depth through a Dense Bundle Adjustment layer to achieve highly accurate and robust performance. Compared to previous work, DROID-SLAM experiences substantially fewer catastrophic failures and offers significant improvements. Notably, the system can leverage stereo or RGB-D video during test time even though it has been trained on monocular video, allowing for enhanced performance in various scenarios. The researchers have made their open-source code available on GitHub at https://github.com/princeton-vl/DROID-SLAM. Overall, DROID-SLAM represents a major advancement in the field of visual SLAM systems. Its deep learning-based approach combined with recurrent iterative updates and Dense Bundle Adjustment layer enables exceptional accuracy and robustness. With its ability to utilize stereo or RGB-D video for improved performance, DROID-SLAM holds great promise for applications requiring precise localization and mapping capabilities.