基于知识学习的多旋翼飞行机器人非零初始状态控制方法研究

Multi-rotor flying robot, launched by aircraft, probing and sensing applications have been deployed. However, due to systematic nonlinearity and highly coupled dynamic natural, its stability and safety envelope is well related to its initial states, which means it is unable to achieve full-state launching stability and trajectory guided control based on the aerodynamic model. Thus, the applications are limited. The study of multi-rotor flying robot non-zero initial states launching control problem is significant valuable to both practical and theoretical aspects. This project will analyze the knowledge decision mathematic model of aircraft human control and propose a general control mechanism of multi-rotor flying robot based on knowledge learning to solve the limited applicability inherent with non-zero initial states. Firstly, human decision behavior during the flight operation will be analyzed and its decision process will be modeled mathematically; then based on this model, combined with uncertain boundary online estimation and reverse enhanced learning methods, the stabilization and trajectory guided control approach of multi-rotor flying robot will be proposed. Therefore, this mechanism merges the advantage of human decision with dynamic control methods to improve the controllability of multi-rotor flying robot with non-zero initial states. This project is a meaningful exploration of aerial robotic control theory as well as strong support of multi-rotor flying robot launching and sensing application in theoretical perspective.

飞行器挂载具有对称结构的多旋翼飞行机器人进行"投放式探测"应用已经展开。然而由于多旋翼飞行机器人非线性与强耦合的动力学特性，其稳定性和安全包线与初始状态密切相关，无法基于动力学模型完成任意飞行状态投放时的镇定与轨迹跟踪控制，限制了其应用范围。对于非零初始速度和姿态下的多旋翼飞行机器人控制方法研究具有重要的理论意义和应用价值。本项目将分析人在飞行器操纵中的知识决策数学模型，提出基于知识学习的多旋翼飞行机器人共性控制方法，以解决非零初始状态控制的科学问题。首先对遥控操作中人的行为进行分析并建立知识决策过程的数学模型描述；然后基于决策模型，通过不确定界在线估计和逆向强化学习方法，提出多旋翼飞行机器人镇定与轨迹跟踪控制方法，将人决策的优势与常规动力学控制方法相结合，提升多旋翼飞行机器人非零初始状态的控制能力。本课题既是对机器人学习理论一次有益探索，又为多旋翼飞行机器人投放式探测应用提供理论支撑。

本项目针对飞行器挂载具有对称结构的多旋翼飞行机器人进行"投放式探测"应用开展研究。由于多旋翼飞行机器人非线性与强耦合的动力学特性，其稳定性和安全包线与初始状态密切相关，无法基于动力学模型完成任意飞行状态投放时的镇定与轨迹跟踪控制，限制了其应用范围。本项目的研究对于非零初始速度和姿态下的多旋翼飞行机器人控制方法研究具有重要的理论意义和应用价值。项目实施过程中分析了人在飞行器操纵中的知识决策数学模型，提出基于知识学习的多旋翼飞行机器人共性控制方法，以解决非零初始状态控制的科学问题。首先对遥控操作中人的行为进行分析并建立知识决策过程的数学模型描述；然后基于决策模型，通过不确定界在线估计和逆向强化学习方法，提出多旋翼飞行机器人镇定与轨迹跟踪控制方法，将人决策的优势与常规动力学控制方法相结合，提升多旋翼飞行机器人非零初始状态的控制能力。本课题既是对机器人学习理论一次有益探索，又为多旋翼飞行机器人投放式探测应用提供理论支撑。