Abstract: The working conditions of construction machinery are complex and the working environment is harsh. The realization of unmanned driving of construction machinery will bring significant social benefits and economic value. Construction machinery is different from ordinary passenger vehicles. In order to achieve high robust localization of driverless construction machinery under multiple complex working conditions, a multi-sensor tight coupled simultaneous localization and mapping (SLAM) system is used. Aiming at the problems of computational redundancy and low computational efficiency of the existing SLAM front-end algorithm, a method based on point line and point surface feature matching is adopted, combined with local map registration, which effectively reduces the amount of data on point cloud registration and avoids computational redundancy. Aiming at the problem of loopless detection of laser fusion SLAM, the loop closure detection mechanism is introduced into the SLAM system based on the spatial nearest neighbor principle and the normal distribution transformation algorithm, which effectively reduces the global error of the SLAM system mapping. Aiming at the problem of easy degradation in the localization and mapping of construction machinery working environment, based on the matching of point line and point surface features, a residual adaptive feedback mechanism is established, which makes the iterative equation solution converge quickly without divergence. The simulation and real vehicle test results show that this SLAM system can effectively solve the problem of map building degradation of construction machinery in bridge, corridor and other operating scenarios, and the speed of map building is greatly improved, which can meet the requirements of real-time localization and mapping of construction machinery.

Key words: unmanned driving, construction machinery, simultaneous localization and mapping, feature matching, loop closure detection, residual adaptive feedback