VAR-SLAM introduces a novel feature-based visual SLAM system designed for dynamic environments. It combines a lightweight semantic keypoint filter to deal with known moving objects, with Barron’s adaptive robust loss to handle unknown ones. An adaptive parameter α is estimated online, allowing the system to handle both static and dynamic regions effectively without extensive threshold tuning.
The paper is available here!!
The implementation will be released publicly after the review process.
Our code builds on ORB-SLAM3.
This repository is maintained by João Soares and Gabriel Fischer.