Mechanical Engineering (B.S.)
Document Type and Release Option
Thesis (open access)
Fully autonomous vehicles must accurately estimate the extent of their environment as well as their relative location in their environment. A popular approach to organizing such information is creating a map of a given physical environment and defining a point in this map representing the vehicle’s location. Simultaneous Mapping and Localization (SLAM) is a computing algorithm that takes inputs from a Light Detection and Ranging (LiDAR) sensor to construct a map of the vehicle’s physical environment and determine its respective location in this map based on feature recognition simultaneously. Two fundamental requirements allow an accurate SLAM method: one being accurate distance measurements and the second being an accurate assessment of location. Researched are methods in which a 2D LiDAR sensor system with laser range finders, ultrasonic sensors and stereo camera vision is optimized for distance measurement accuracy, particularly a method using recurrent neural networks. Sensor fusion techniques with infrared, camera and ultrasonic sensors are implemented to investigate their effects on distance measurement accuracy. It was found that the use of a recurrent neural network for fusing data from a 2D LiDAR with laser range finders and ultrasonic sensors outperforms raw sensor data in accuracy (46.6% error reduced to 3.0% error) and precision (0.62m std. deviation reduced to 0.0015m std. deviation). These results demonstrate the effectiveness of machine learning based fusion algorithms for noise reduction, measurement accuracy improvement, and outlier measurement removal which would provide SLAM vehicles more robust performance.
Mehrzed, Shaen, "Optimization of a Simultaneous Localization and Mapping (SLAM) System for an Autonomous Vehicle Using a 2-Dimensional Light Detection and Ranging Sensor (LiDAR) by Sensor Fusion" (2023). Honors College Theses. 861.