Nonlinear Model Predictive Path-Following Controller for a Small-Scale Autonomous Bulldozer for Accurate Placement of Materials and Debris of Masonry in Construction Contexts

Abstract

This paper presents a nonlinear model predictive control (NMPC) scheme for the medium-level control of a small-scale autonomous bulldozer to accurately and safely displace crushed materials of masonry in construction contexts. For this purpose, the controller is required to minimize the error between the achieved and required paths; additionally, the control actions must be smooth for minimizing the mistreatment of equipment, which is intended to operate in the long term, as it is usual in industrial-grade solutions. The proposed NMPC based path-follower can adequately handle the platform's constraints and usual perturbations. In terms of state estimation, a map-based localizer is implemented via an extended Kalman filter (EKF) based on the platform nominal process model and light detecting and ranging (LiDAR) and inertial measurement unit (IMU) sensors measurements. The localizer provides estimates of the platform's pose, necessary for the MPC controller's state feedback. An actual experiment on a modified UGV (Clearpaths Husky A-200) is performed to validate the performance of the proposed control scheme. The UGV is retrofitted with a blade (for pushing material) and appropriate sensors for the necessary perception tasks. Experimental results indicate that the proposed control scheme is robust and suitable for safely pushing the crushed materials, presenting appropriately low deviation from the nominal path and requiring reasonably low processing time.