Rescue Robots

• Tech Stack: C++, ROS, SLAM

In this project, I was a part of Mapping Subgroup. This subgroup is concerned about improving the existing map quality. This is achieved by enhancing the robot's localization, trying different optimization techniques, increasing the reliability of loop closure and filtering dynamic objects within the mapping system. The project builds on the LOAM (Lidar Odometry and Mapping) algorithm, incorporating IMU data and loop closure for enhanced map accuracy. A templated structure was created to support multiple graph optimizers, specifically g2o and SE-Sync, allowing for flexible and efficient optimization. It also implemented a dynamic object filter using SegFormer B1 for semantic segmentation, which projects 3D Lidar points onto a 2D image plane to detect and remove dynamic objects, improving map quality. Various test cases were conducted to assess the impact of IMU calibration, loop closing, and dynamic object filtering on map quality. Results showed significant improvements with calibrated IMU data and enabled loop closing.