Cooperative Fault-Tolerant Formation Tracking Control for Heterogeneous Air-Ground Systems Using a Learning-Based Method

This paper investigates the formation tracking control problem for heterogeneous air-ground systems with quadrotor unmanned aerial vehicles (UAVs) and wheeled unmanned ground vehicles (UGVs) of nonlinear dynamics. A hierarchical learning-based framework is proposed to address the unknown tracking leader and multiple faults. First, an adaptive observer is designed to estimate both the dynamics and state of the tracking leader via a directed network. Second, based on the data from an internal model filter and system measurements, an off-policy reinforcement learning (RL) algorithm is applied to learn a unified tracking planner for both UAVs and UGVs. The planner is coordinated with time-varying leader and formation in the RL process, while internal dynamics are learned. Third, a finite-time differential filter and a virtual reference are respectively constructed for UAVs and UGVs, and transform the planner's outputs into tracking commands. By integrating adaptive neural networks and robust compensations with backstepping designs, the fault-tolerant formation tracking is realized. Finally, the learning and control framework on air-ground systems are validated by numerical simulations.