全方位机器人的重心位置预测与轨迹跟踪控制

Center of gravity position prediction and trajectory tracking control for omnidirectional robots

中文关键词  跟踪控制  时变矩阵求逆  重心偏移  参数估计  长短期记忆神经网络

英文关键词  tracking control  inverse time-varying matrix  center of gravity offset  parameter estimation  long short-term memory

中文摘要

      针对全方向移动机器人存在非线性动态强耦合、实时重心偏移及难以实现高精度跟踪控制的问题, 本文提 出一种基于长短期记忆(LSTM)神经网络的重心位置在线预测的轨迹跟踪控制法. 首先, 建立考虑重心偏移的动力 学模型并基于LSTM神经网络训练构建其对比模型; 其次, 基于模型对比法实时估计重心偏移参数, 再基于张神经 网络(ZNN)对估计的重心偏移参数进行预测以减小估计过程引起的滞后; 最后, 基于非线性动态反馈解耦法设计数 值加速度控制算法, 且基于离散系统极点配置法分析了系统的稳定性. 仿真结果验证了所提方法相对于数值加速 度控制器与自适应控制器因能在线预测重心偏移参数完成高精度动态解耦实现控制精度的提高. 实际实验中, 所 提控制算法相比数值加速度控制及模型预测控制, 其跟踪精度明显提高, 这表明所提控制算法可显著减小重心偏移 对跟踪控制精度的影响.

英文摘要

      A tracking control method based on the long short-term memory (LSTM) neural network for on-line prediction of the position of the center of gravity of an omnidirectional mobile robot is presented to solve the problems of nonlinear dynamic strong coupling, real-time center of gravity offset and difficulty in achieving high-precision tracking control. Firstly, a dynamic model considering gravity center deviation is established and its contrast model is built based on the LSTM neural network training. Secondly, the center of gravity offset parameters are estimated in real time based on the model comparison method, and then the center of gravity offset parameters are predicted based on the Zhang neural network (ZNN) to reduce the lag caused by parameter estimation. Finally, a numerical acceleration control algorithm is designed based on the dynamic feedback decoupling method, and the stability of the system is analyzed based on the pole assignment method of discrete system. The simulation results verify that the proposed method can improve the control accuracy by high-precision dynamic decoupling compared with the numerical acceleration controller and the adaptive controller because of the ability to predict the center of gravity offset parameters online. In actual experiments, the tracking accuracy of the proposed control algorithm is significantly higher than that of numerical acceleration control and model predictive control, which indicates that the proposed control algorithm can significantly reduce the impact of center of gravity offset on tracking control accuracy.