A new metric for assessing the performance of 2D Lidar SLAMs

Simultaneous Localisation and Mapping (SLAM) is a widely studied topic in recent years and has a wide potential in the field of unmanned driving and robotics. Over the past decade, a number of SLAM algorithms have been developed, each exhibiting unique performance in their applications. This paper presents a comparative study of the performance of three well-known Light De-tection and Ranging (LiDAR)-based SLAM algorithms, i.e. Gmapping, Cartog-rapher and Hector, with an emphasis on the 2D maps constructed by each algo-rithm. In order to deal with incomplete maps constructed, a new evaluation met-ric was proposed. To reduce the human error during scene construction and equipment calibration, all experiments were carried out in the 2D simulation available within the Robot Operating system (ROS). Three well-designed maps with different sizes and complexities were introduced to investigate the features of three SLAM algorithms. Besides, to reduce the impact of randomness, each dataset was assessed 10 times to obtain the mean value and the standard devia-tion. The results show that, in comparison to traditional metrics such as a metric of average distance to nearest neighbour (ADNN), the proposed measurement can clearly reflect both the quality and completeness of maps built by SLAM algorithms.