Robust Control Policy Transfer between Quadrotors

Abstract

Since their advent, quadrotors (QUAVs) have been widely used in various tasks. Control algorithms for quadrotors are designed and tuned based on performing specific tasks. However, when a new quadrotor with different dynamic parameters, such as mass and moment of inertia, is required to perform the same task, the baseline controller must be modified, or at least tuned again, to achieve the same performance level. Hence, knowledge transfer for tracking the desired path between QUAVs with different dynamics without changes in the baseline controller is desirable. In this regard, extensive efforts have been made to provide autonomous control approaches to guarantee QUAV tracking performance without changing the baseline controller design. The idea of automatically improving the tracking performance of a quadrotor, based on feeding it with the tracking errors and experiences collected from another quadrotor with different dynamics, is a challenging issue. In this thesis, a robust control policy transfer algorithm from a so-called source quadrotor to a target quadrotor with different dynamic parameters is proposed. First, a robust controller is designed to control the source quadrotor with both H and µ synthesis approaches by taking into consideration parametric uncertainties, disturbances, and noise. Then a robust control policy transfer algorithm is designed for transferring the trajectory tracking experience from the source quadrotor to the target one where the basic controllers of each QUAV are designed based on H. In the proposed approach, the signal generated by the control policy block from the source QUAV is added to the feedforward loop of the target QUAV. Additionally, a genetic algorithm (GA) is utilized to optimize the parameters of the control policy transfer block, ensuring the design meets stability requirements. It is worth mentioning that the control policy transfer block is exploited as an additional algorithm to further improve the QUAV tracking performance and does not affect closed-loop stability. Simulation-based studies show that the suggested algorithm improves the tracking performance for the target quadrotor without modifying the baseline controller while still guaranteeing robustness. Furthermore, a parameter sensitivity study was conducted with quadrotors having varying masses and moments of inertia to evaluate the robustness of the control policy transfer algorithm. The simulations showed that the proposed algorithm consistently enhanced the tracking performance of the target quadrotor, even under significant parametric variations. The best tracking results were observed with the nominal target quadrotor, while the quadrotor with a 20% mass reduction exhibited minor overshoot due to increased system sensitivity. Despite these variations, the algorithm maintained robust stability, with gain margins exceeding 30 dB and adequate phase margins across all control channels. These results confirm the algorithm's capability to preserve stability and robust performance across a range of dynamic conditions.