Research And Implementation Of Self-Adaptive Multi-Sensor Fusion Simultaneous Localization And Mapping Technology

Pose estimation using on-board sensors are critical for mobile robots.Due to the different performance of different sensors in different environments,it is a challenge to fuse the information of various sensors to adapt to various environments.This paper proposes a tightly coupled multi-sensor fusion SLAM framework,Lvio-Fusion,which integrates four sensors:stereo camera,lidar,inertial measurement unit and satellite positioning based on graph optimization,and supports the expansion of more types of sensors.Especially for urban traffic scenarios,the framework introduces satellite positioning and closed-loop constrained segmented global pose graph optimization,which can eliminate accumulated drift.Furthermore,the framework creatively uses the actor-critic method in reinforcement learning to allow the algorithm to adaptively adjust the sensor weights.After training,the adaptive algorithm can provide a better sensor factors’weights for the system.This paper evaluates the performance of the system on public datasets and compares it with other state-of-the-art methods,showing that the proposed method has high estimation accuracy and robustness for various environments.And the implementation of the algorithm is open source and highly extensible.This work can be summarized into the following points:(1)Realize a tightly coupled multi-sensor fusion framework based on optimized stereo camera,lidar,inertial measurement unit and satellite positioning to achieve high-precision,real-time trajectory estimation of mobile robots;(2)A segmented global optimization method for urban traffic scenes,which can effectively eliminate accumulated drift and provide global location.(3)Using an adaptive algorithm based on the actor-critic method in reinforcement learning,the weight of the sensor can be adaptively adjusted for different environments.In order to achieve the adaptability of complex and diverse scenes.