Task-Driven SLAM Benchmarking for Dynamic Indoor Navigation in ROS2

Current simultaneous localization and mapping (SLAM) systems are crucial for autonomous robotic applications but have been proven to face difficulties in dynamic indoor environments, where moving objects can lead to inaccurate localization and navigation errors. These errors can result in flawed maps, incorrect path planning, and robot collisions. Although static environments are well supported, there is limited benchmarking of SLAM algorithms for dynamic environments. This paper addresses these gaps by advancing a dynamic, task-driven benchmarking framework for evaluating SLAM algorithms within a Robot Operating System (ROS2). Task-driven metrics emphasize consistency, such as pose variance between tests, success rates, and average duration. Through simulation tests, key discrepancies are highlighted within algorithms’ localization and navigation around dynamic objects. After conducting benchmarks on a Visual SLAM and a 2D Lidar SLAM implementation, it was determined that Visual SLAM implementations have greater precision across task-driven experiments, while 2D Lidar SLAM has greater success reliability.