Jetson-SLAM

Published at IEEE Robotics and Automation Letters 2023

Presented at Internation Conference on Robotics and Automation 2024 (ICRA)

More updates coming soon

To be done: 1. Update exec folder

The frontend of Jetson-SLAM will be released as a separate repository as-well.

Authors: Ashish Kumar

This repository contains Jetson-SLAM with FULL-BA back-end Jetson-SLAM is a GPU-thrusted real-time SLAM library for Monocular, Stereo and RGB-D cameras. It can run very high speeds beyond 500FPS on RTX-2070 and Beyond 90FPS on Jetson-NX @320x240 resolution. Please see the Jetson-SLAM Paper for rigorous results over different resolutions, GPUs and comparison with existing VO/VIO/SLAM pipelines.

Jetson-SLAM can run alongside Deep Neural Networks. It is fully behchmarked with VGG

Video

Main Highlight

1. Main Results

Datasets

1. KITTI-Benchmark

2. EuRoC Benchmark

3. KAIST-VIO Benchmark

Results on KITTI Benchmark

Results on EuRoC Benchmark

Results on KAIST-VIO Benchmark

Performance with scaled versions of VGG-16 Co-existing on Jetson-NX

Build Instructions

**Step-1**
 Install the dependencies given below:

1. OpenCV 3
2. Eigen3
3. CUDA
4. Pangolin


**Step-2**
 Run build.sh

Run Instructions

Go to execs and run Jetson-SLAM on following choices:
1. Run stereo_kitti for KITTI-Benchmark
2. Run stereo_euroc for EuRoC Benchmark
3. Run stereo_kaistvio for KAIST-VIO Benchmark
4. Run stereo_live for live images from a Stereo-Rig. Please customize the "stereo_live_config.yaml" file for your stereo rig.

License

Jetson-SLAM is released under a [GPLv3 license].

Bibtex citation:

 @article{Jetson-SLAM,
   title={{High-Speed Stereo Visual-{SLAM} For Low-Powered Computing Devices},
   author={Kumar, Ashish and Park, Jaesik and Behera, Laxmidhar},
   year={2023},
  }