针对传统逆运动问题数值求解算法计算时间长,多解情况下计算结果单调的问题,采用回复式神经网络对逆运动问题进行近似优化求解.通过针对逆运动问题建立基于回复式神经网络的动态求解模型,采用神经元异步更新计算和初始状态扰动,保证多解情况下解运动迹线的多样性.文中算法在无解情况下有较好适应性和稳定性,能迅速收敛到近似最优状态.算法理论计算时间复杂度为O(n2),可满足实时应用的需求.实验对典型对子运动链运动系统进行模拟,结果表明算法具有稳定和收敛性.%The conventional Inverse Kinematics problem solving methods have many disadvantages such as slowly convergence speed or monotonous computation result.This issue presents a novel algorithm with recurrent neuron-networks model to treat the Inverse Kinematics problem as approximate optimization solving.This method designs proper optimization function for Inverse Kinematics system and utilizes recurrent neuron-networks computational model to resolve.During the computation,the neurons were updated asynchronously and the initial system state was disturbed to obtain diverse trajectories when multi-solutions exist.The algorithm has good stability and convergence ability even when no solution exists.The theoretical computation time complexity of the algorithm is O(n2),and can meet general real-time application requirements.Experiments were done to simulate the dyad kinematics system.The calculation results show that the computational time are increase with the neuron number but not exponentially,that are in good agreement with the algorithm convergence stability.
展开▼
