Research On SLAM Algorithm Of Mobile Robot Based On Multi-sensor Fusion

Robot technology can not only reflect a country’s development capabilities and intelligent level,but also is one of the important ways for a country to become strong.Simultaneous localization and mapping(SLAM)is one of the basic technologies for mobile robots to complete various complex tasks.SLAM with a single type of sensor has been widely used in statically stable and uncomplicated environments.However,for some complex indoor scenes,a single sensor SLAM cannot meet the needs of such scenes due to its own shortcomings.Therefore,combining complementary sensors to construct a multi sensor fusion based SLAM has certain practical significance and practical value.For complex indoor environments,this thesis proposes a mobile robot SLAM based on multi-sensor fusion to enhance the localization accuracy and mapping quality of the SLAM in complex indoor environments.The main content of this thesis includes the following aspects:(1)Constructing a position and orientation estimation algorithm for mobile robots based on multi-sensor fusion.Firstly aiming at the problem of poor discrimination ability of laser pose estimation algorithms based on scanning matching in weak feature environments,a wheel odometry is added to the scanning matching of laser to build a laser-wheel odometry to enhance the working ability of laser in different environments;Secondly aiming at the problem that visual-inertial odometry(VIO)are not suitable for mobile robots,a laser-wheel odometry is introduced into the VIO framework based on multi-state constraint Kalman filter to construct a robust and stable pose estimation algorithm for mobile robots;Finally experiments on pose estimation of mobile robots on open source datasets show that the proposed pose estimation algorithm based on multisensor fusion has certain robustness and localization accuracy.(2)Constructing a global optimization algorithm based on multi-sensor fusion.Aiming at the problem of error accumulation in long time and wide range SLAM,this thesis first selects key frames for loop closure detection;Then,a graph optimization model based on multi-sensor fusion is constructed to estimate the robot’s posture and the location of spatial feature points;Then,using the optimized results to construct a twodimensional grid map,and combining spatial feature points to form a multi-source map;Finally,experiments were conducted on the optimization algorithm and mapping method on an open source dataset.The experimental results show that the optimized robot pose error is effectively reduced,and the multi-source map generated by the experiment can effectively restore some features of the robot work scene.(3)Building a robot experimental platform to validate the algorithm proposed in this thesis.Firstly,a hardware and software platform for robot experiments is built,and then the internal and external parameters of each sensor are calibrated.Finally,the algorithm studied in this thesis is validated in laboratory and corridor scenarios.The experimental results show that the SLAM studied in this thesis has certain advantages over common open source SLAM in localization accuracy and mapping effects in complex indoor scenes.