sair-lab/AirSLAM

Health v2 · maintenance only

Adoption not part of the score

1190 stars · 176 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

AirSLAM is an efficient, illumination-robust point-line visual SLAM system supporting stereo visual odometry/VIO, offline map optimization, and lightweight relocalization. It combines a unified CNN for keypoint and line detection (accelerated with C++ and NVIDIA TensorRT) with traditional geometric backend optimization.

Use cases

• run visual SLAM in illumination-challenging environments
• estimate camera trajectory from stereo or stereo-inertial video
• relocalize a robot against a previously built map
• deploy real-time SLAM on embedded platforms like Jetson
• build and optimize point-line maps offline
• research deep-learned feature detection and matching for SLAM

When to choose

• you need robust visual odometry/SLAM under lighting changes
• you want GPU-accelerated SuperPoint/SuperGlue-style features in C++ with TensorRT
• you need high frame rates (40-73Hz) on PC or embedded hardware
• you need both V-SLAM and VI-SLAM modes with relocalization

When to avoid

• you need LiDAR or RGB-D based SLAM
• you need a pure CPU solution without NVIDIA GPUs
• you need a plug-and-play library with a stable API rather than a research system
• your project is incompatible with the GPL-3.0 license

Facets

library · maturity active

computer-vision machine-learning simulation robotics computer-vision autonomous-vehicles cpp visual-slam visual-odometry superpoint superglue tensorrt relocalization point-line-features vslam vio linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="sair-lab/AirSLAM")