Active Fault-Tolerant Control Based on MPC and Reinforcement Learning for Quadcopter with Actuator Faults

Huicheng Jiang,Feng Xu,Xueqian Wang,Songtao Wang
DOI: https://doi.org/10.1016/j.ifacol.2023.10.589
2023-01-01
IFAC-PapersOnLine
Abstract:In this paper, we propose an active fault-tolerant control (AFTC) method combining model predictive control (MPC) and reinforcement learning (RL) for the quadcopter with actuator faults. We take a data-based discriminant model as the fault detection and diagnosis (FDD) module indicating a system fault mode based on the state error. With the information of the fault mode and the state error, the RL controller generates auxiliary control signals to correct the system. To configure the MPC controller quickly, we propose an auxiliary signal-based method for estimation of fault parameters and prove its convergence. The AFTC framework reduces requirements for accurate modeling, and avoids the instability of the RL controller under a continuous operation. To validate the effectiveness of the proposed framework, two trajectory tracking simulations with single and multiple faults are carried out. The simulation results show satisfactory performance and verify that the proposed framework is real-time applicable.
What problem does this paper attempt to address?