Research On Tracking Control And Trajectory Planning Of Quadrotor With Multi Constraints

In order to improve the autonomous flight ability of quadrotor, this paper study on its flight control problem with multiple constraints under complex environments from two different perspectives. From tracking control perspective, model predictive control theory and deep-learning technique are employed to propose a tracking controller design method and its implementation scheme which meet multiple constraints. From trajectory planning perspective, based on RRT algorithm and model prediction approach, a trajectory planning scheme is designed to meet multiple constraints. The main contributions can be summarized as follows:First, the modelling of quadrotor is carried out. And using the Taylor expansion linearization method, the nonlinear model of the aircraft is approximated as a linear time varying model, which facilitates the following research. The common constraints of quadrotor are summarized to obtain the general expressions, and the problems of tracking control and trajectory planning with multiple constraints are presented.Secondly, the model-predictive-control-based design method of trajectory tracking controller of quadrotor aircraft and the deep-neural-network-based realization method of controller are proposed. First, using model separation method, the system is divided into two parts, i.e. the attitude subsystem and the position subsystem, to reduce the complexity of control design. Then, giving consideration to the tracking error and the input change rate, the corresponding performance index function is constructed. Furthermore, taking the state and input constraints into account, model predictive control approach is applied to transform the constrained optimization problem into QP(Quadratic Program) problem. To reduce the large calculation amount during solving numerical optimization, the deep neural network is used as actual controller instead of the original one after it is trained by the QP solver. This strategy not only guarantees the MPC performance, but also improves the algorithm in real-time greatly. Extensive simulations verify the effectiveness and superiority of the strategy.Last, a trajectory planning strategy for quadrotor based on RRT(Rapidly-exploring Random Trees) and model prediction approach is proposed. Firstly, the RRT algorithm is employed to implement the path searching. Considering poor repeatability and unflatness of the trajectories generated by existing RRT algorithm, a modified RRT algorithm is proposed by using adjusted probability space and time distribution and expansion-deletion technique to obtain more ideal path sequence while maintaining the barrier ability. Then, the polynomial fitting and the model prediction approaches are used to realize the trajectory fitting. Polynomial-fitting-based method can solve the terminal state constraints, and the model-prediction-based method combining with rolling optimization can solve both terminal constraints and process constraints. The simulation results verify the effectiveness of the proposed trajectory planning scheme.Although the research in this paper focuses on the quadrotors, those proposed strategies and methods, in principle, are also suitable for other aircrafts, even for other similar systems such as ground robots and underwater robots. The results may provide helpful references for solving tracking control and trajectory planning problems of such systems with multiple constraints.