[目的]为解决非结构化环境下采用深度强化学习进行采摘机械臂路径规划时存在的效率低、采摘路径规划成功率不佳的问题,提出了一种非结构化环境下基于深度强化学习(Deep reinforcement learning, DRL)和人工势场的柑橘采摘机械臂的路径...[目的]为解决非结构化环境下采用深度强化学习进行采摘机械臂路径规划时存在的效率低、采摘路径规划成功率不佳的问题,提出了一种非结构化环境下基于深度强化学习(Deep reinforcement learning, DRL)和人工势场的柑橘采摘机械臂的路径规划方法。[方法]首先,通过强化学习方法进行采摘路径规划问题求解,设计了结合人工势场的强化学习方法;其次,引入长短期记忆(Longshort term memory,LSTM)结构对2种DRL算法的Actor网络和Critic网络进行改进;最后,在3种不同的非结构化柑橘果树环境训练DRL算法对采摘机械臂进行路径规划。[结果]仿真对比试验表明:结合人工势场的强化学习方法有效提高了采摘机械臂路径规划的成功率;引入LSTM结构的方法可使深度确定性策略梯度(Deep deterministic policy gradient,DDPG)算法的收敛速度提升57.25%,路径规划成功率提升23.00%;使软行为评判(Soft actor critic,SAC)算法的收敛速度提升53.73%,路径规划成功率提升9.00%;与传统算法RRT-connect(Rapidly exploring random trees connect)对比,引入LSTM结构的SAC算法使规划路径长度缩短了16.20%,路径规划成功率提升了9.67%。[结论]所提出的路径规划方法在路径规划长度、路径规划成功率方面存在一定优势,可为解决采摘机器人在非结构化环境下的路径规划问题提供参考。展开更多
This paper presents the fundamentals of a continuous adjoint method and the applications of this method to the aerodynamic design optimization of both external and internal flows.General formulation of the continuous ...This paper presents the fundamentals of a continuous adjoint method and the applications of this method to the aerodynamic design optimization of both external and internal flows.General formulation of the continuous adjoint equations and the corresponding boundary conditions are derived.With the adjoint method,the complete gradient information needed in the design optimization can be obtained by solving the governing flow equations and the corresponding adjoint equations only once for each cost function,regardless of the number of design parameters.An inverse design of airfoil is firstly performed to study the accuracy of the adjoint gradient and the effectiveness of the adjoint method as an inverse design method.Then the method is used to perform a series of single and multiple point design optimization problems involving the drag reduction of airfoil,wing,and wing-body configuration,and the aerodynamic performance improvement of turbine and compressor blade rows.The results demonstrate that the continuous adjoint method can efficiently and significantly improve the aerodynamic performance of the design in a shape optimization problem.展开更多
文摘[目的]为解决非结构化环境下采用深度强化学习进行采摘机械臂路径规划时存在的效率低、采摘路径规划成功率不佳的问题,提出了一种非结构化环境下基于深度强化学习(Deep reinforcement learning, DRL)和人工势场的柑橘采摘机械臂的路径规划方法。[方法]首先,通过强化学习方法进行采摘路径规划问题求解,设计了结合人工势场的强化学习方法;其次,引入长短期记忆(Longshort term memory,LSTM)结构对2种DRL算法的Actor网络和Critic网络进行改进;最后,在3种不同的非结构化柑橘果树环境训练DRL算法对采摘机械臂进行路径规划。[结果]仿真对比试验表明:结合人工势场的强化学习方法有效提高了采摘机械臂路径规划的成功率;引入LSTM结构的方法可使深度确定性策略梯度(Deep deterministic policy gradient,DDPG)算法的收敛速度提升57.25%,路径规划成功率提升23.00%;使软行为评判(Soft actor critic,SAC)算法的收敛速度提升53.73%,路径规划成功率提升9.00%;与传统算法RRT-connect(Rapidly exploring random trees connect)对比,引入LSTM结构的SAC算法使规划路径长度缩短了16.20%,路径规划成功率提升了9.67%。[结论]所提出的路径规划方法在路径规划长度、路径规划成功率方面存在一定优势,可为解决采摘机器人在非结构化环境下的路径规划问题提供参考。
基金supported by the National Natural Science Foundation of China(Grant Nos.51206003 and 51376009)the National Science Foundation for Post-doctoral Scientists of China(Grant Nos.2012M510267 and 2013T60035)
文摘This paper presents the fundamentals of a continuous adjoint method and the applications of this method to the aerodynamic design optimization of both external and internal flows.General formulation of the continuous adjoint equations and the corresponding boundary conditions are derived.With the adjoint method,the complete gradient information needed in the design optimization can be obtained by solving the governing flow equations and the corresponding adjoint equations only once for each cost function,regardless of the number of design parameters.An inverse design of airfoil is firstly performed to study the accuracy of the adjoint gradient and the effectiveness of the adjoint method as an inverse design method.Then the method is used to perform a series of single and multiple point design optimization problems involving the drag reduction of airfoil,wing,and wing-body configuration,and the aerodynamic performance improvement of turbine and compressor blade rows.The results demonstrate that the continuous adjoint method can efficiently and significantly improve the aerodynamic performance of the design in a shape optimization problem.
