Fundamental scan matching approach to LIDAR based localization

Joel Douglas Larson

doi:10.7273/000004210

Autonomous robotics is a rapidly growing field and of increasing importance to undergraduate engineering programs. As technology advances, robots are moving past basic repetitive jobs and finding uses that further expand human capabilities. They are becoming more capable of taking over dangerous tasks, as well as freeing humans up for more meaningful work. This progress requires robots to be able to sense their environment and act without supervision. Toward this end, a major problem in autonomous robotics is being addressed, which is simultaneous localization and mapping (SLAM). SLAM is still very much in its research and development phase. Currently there are many methods available which to varying degrees solve the localization problem of SLAM. The majority use probabilistic concepts. However, these solutions often bring with them a high level of algorithmic complexity. This can raise the cost of implementation and make fundamental concepts difficult to comprehend, which is especially a problem for engineering students being exposed to SLAM principles for the first time. Furthermore, because high complexity may not be necessary when the constraints of time and accuracy are not as severe, some applications do not require the rigor of a full probabilistic SLAM method. Therefore, a solution to the localization problem which focuses on clarity and first principles can be beneficial to the engineering-education community and for applications where low cost is favored over high performance. The aim of this study is to develop a practical and relatively simple robotic development platform as a starting point for future research and development. The platform and research results will also be incorporated into a laboratory course covering the fundamentals of localization. This work presents a straightforward approach to the localization portion of the SLAM problem using scan matching techniques. By reducing scan matching from a multi-dimensional to a one-dimensional problem, localization can be solved in an easily understandable way while minimizing cost and computational requirements. This solution provides estimates of the robot pose within the confines of a static/simple environment. The techniques utilized are cross correlation for rotation movements combined with simple root mean squared error calculations for translation movements.

Fundamental scan matching approach to LIDAR based localization

Files and links (1)

Abstract

Metrics

Details