面向复杂未知多障碍环境的多无人机分布式在线轨迹规划  

Multi-UAV decentralized online trajectory planning in complex unknown obstacle-rich environments

作　　者：张学伟 田栢苓 鲁瀚辰 谌宏鸣 宗群 Xuewei ZHANG;Bailing TIAN;Hanchen LU;Hongming SHEN;Qun ZONG(School of Electrical and Information Engineering,Tianjin University,Tianjin 300072,China;Science and Technology on Complex System Control and Intelligent Agent Cooperation Laboratory,Beijing Electro-Mechanical Engineering Institute,Beijing 100074,China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津300072 [2]北京机电工程研究院复杂系统控制与智能协同技术重点实验室,北京100074

出　　处：《中国科学：信息科学》2022年第9期1627-1641,共15页Scientia Sinica(Informationis)

基　　金：国家优秀青年科学基金(批准号:62022060);国家自然科学基金面上项目(批准号:62003236,61873340,61903349)资助。

摘　　要：考虑复杂未知多障碍环境对无人机实时轨迹规划性能的影响,提出了基于Tube-MPC和模型预测路径积分(model predictive path integral,MPPI)控制相结合的多无人机分布式实时轨迹规划框架与方法.首先,考虑无人机在多障碍环境下的避碰避障需求,构造代价函数表征轨迹规划过程中的约束条件,将多无人机的轨迹规划问题转化为随机最优控制问题.其次,借鉴Tube-MPC思想,设计并实现了多无人机分布式轨迹规划框架,通过将低频标称控制器与高频辅助控制器串联保证了系统的实时性和鲁棒性.再次,为避免传统方法在求解过程中的维数灾难,提出基于MPPI的多无人机异步轨迹规划方法,该方法通过基于GPU的并行蒙特卡洛(Monte-Carlo)随机前向采样技术,将多无人机随机最优控制问题的求解转化为给定代价函数下对采样轨迹期望的求解,进而获得最优控制序列,其显著特点是求解速度快且避免了基于梯度求解方法对约束条件和代价函数连续性及凸特性的要求.最后,通过Gazebo虚拟仿真平台,在复杂未知多障碍环境下对算法的有效性进行了验证.In consideration of the effect of complex unknown obstacle-rich environments on the performance of the real-time trajectory planning of unmanned aerial vehicles(UAVs),this paper proposes a framework and method for multi-UAV decentralized real-time trajectory planning based on Tube-MPC and model predictive path integral(MPPI)control.First,the cost function representing constraints is formulated by taking into account the requirements of the reciprocal collision avoidance and the obstacle avoidance in an obstacle-rich scenario such that the trajectory planning problem is transformed into a stochastic optimal control problem.Furthermore,a decentralized trajectory planning framework for multiple UAVs is designed and established based on Tube-MPC,wherein the low-frequency nominal controller and the high-frequency auxiliary controller are connected in series to ensure the real-time characteristic and robustness of the system.In addition,an asynchronous multi-UAV trajectory planning method based on MPPI for avoiding dimensional disaster in solution through traditional methods is presented.Through the parallel Monte-Carlo random forward sampling technology based on GPU,the solution of the stochastic optimal control problem for multiple UAVs is transformed into the expectation of sampling trajectory under a given cost function.Then,the optimal control sequence is obtained.Its notable feature is quick solution and avoidance of the requirements of continuity and convex properties of the constraint condition and the cost function that are indispensable for the gradient-based method.Finally,the effectiveness of the algorithm in a complex,unknown,and obstacle-rich environment is verified using the Gazebo virtual simulation platform.

关 键 词：未知多障碍环境 多无人机 在线轨迹规划 模型预测路径积分 GPU并行加速 

分 类 号：V249.1[航空宇航科学与技术—飞行器设计] V279