An Overview Study of Multi-sensor Fusion for Map Building and Localization

Abstract

This paper reviews the application of multi-sensor fusion in simultaneous localisation and map construction (SLAM) technology. With the development of robotics, autonomous driving and virtual reality, there is an increasing demand for precise localisation and map construction.SLAM technology has emerged to solve the problem of autonomous robot localisation and map construction in unknown environments. However, single-sensor SLAM systems have limitations, such as limited sensing capability and susceptibility to noise interference. Multi-sensor fusion SLAM significantly improves the performance and robustness of the system by integrating the advantages of multiple sensors. The multi-sensor fusion SLAM system includes key components such as sensor data reading, front-end visual odometry, back-end optimisation, loopback detection and map building. The sensor data is first optimised to reduce noise and then further processed according to the task requirements. Sensors are categorised into internal sensors (e.g. IMUs and wheeled odometers) and external sensors (e.g. cameras, LIDAR, UWB sensors, etc.). Common data fusion methods include filter-based fusion (e.g., Extended Kalman Filter, Particle Filter), optimisation-based fusion (e.g., Graph Optimisation, Nonlinear Least Squares) and deep learning-based fusion (e.g., Convolutional Neural Network, Recurrent Neural Network). These methods are able to handle different types of sensor data and improve the performance of SLAM systems. Multi-sensor fusion SLAM technology has a wide range of applications in fields such as robot navigation, autonomous driving, virtual reality, geographic information systems and mapping. In the future, the technology will pay more attention to the combination of multimodal fusion and deep learning, optimising the computational efficiency of the algorithms, improving the real-time and robustness of the system, as well as the fusion of cloud SLAM and edge computing, to promote the development and advancement of the related fields.