Decoupling control based on neural network inverse system for path tracking in multi-actuated unmanned ground vehicles

This paper proposes a novel multi-layer path tracking and vehicle dynamics control architecture targeting all-wheel-independently-actuated unmanned ground vehicles (AWIA UGVs). The control strategy addresses cross-coupling effects by integrating direct force control (DFC) into a decoupling controller, establishing a direct relationship between the intended vehicle motions and the direct force vector at the centre of gravity. The DFC framework consists of three layers: (i) the reference generation layer, in which a model predictive controller (MPC) generates the desired vehicle motion states to track the reference trajectory; (ii) the state tracking layer, which is responsible for the decoupling tracking control of the vehicle motion vector, based on a neural network inverse (NNI) system; and (iii) the control allocation layer, which distributes the control inputs among the redundant actuators. The DFC effectiveness is evaluated through simulations with an experimentally validated high-fidelity model of a prototype UGV, and preliminary proof-of-concept experiments. The results showcase the superior performance of the decoupling algorithm in terms of trajectory tracking and body control performance, in comparison with benchmarking state-of-the-art MPC-based path tracking and vehicle dynamics control methods.