A Lidar SLAM based on Improved Particle Filter and Scan Matching for Unmanned Delivery Robot

Abstract

With the rapid development of robotics, intelligent robots have gradually entered human daily life and take responsibility for specific tasks, which brings tremendous convenience to people’s lives. In indoor places such as restaurants, hotels, logistics warehouses, SLAM technology can guide mobile robots to complete delivery tasks. Aiming at solving the problems of particle degradation in the traditional RBPF-based two-dimensional SLAM algorithm, our algorithm firstly proposes an improved particle filter method based on cross-mutation, which changes the information of the small weight particles in the system by designing cross-mutation operators. Thus, a certain amount of small weight particles are retained in the system to increase the diversity of particles. Secondly, our algorithm proposes a scan matching method based on point and line features, which makes full use of the geometric information of the two-dimensional point cloud by designing a new error function. Compared with traditional matching methods that only rely on point features, it has better matching accuracy. We conduct Two-dimensional SLAM experiments in the Gazebo simulation environment and the real natural environment. The results show that the proposed algorithm has higher mapping accuracy than the commonly used two-dimensional SLAM algorithm Gmapping at this stage. Using improved A* algorithm to conduct path planning experiments on the two-dimensional grid map built with our algorithm, the results prove the effectiveness of our algorithm.