为提高多无人船编队系统的导航能力,提出了一种基于注意力机制的多智能体深度确定性策略梯度(ATMADDPG:Attention Mechanism based Multi-Agent Deep Deterministic Policy Gradient)算法。该算法在训练阶段,通过大量试验训练出最佳策略...为提高多无人船编队系统的导航能力,提出了一种基于注意力机制的多智能体深度确定性策略梯度(ATMADDPG:Attention Mechanism based Multi-Agent Deep Deterministic Policy Gradient)算法。该算法在训练阶段,通过大量试验训练出最佳策略,并在实验阶段直接使用训练出的最佳策略得到最佳编队路径。仿真实验将4艘相同的“百川号”无人船作为实验对象。实验结果表明,基于ATMADDPG算法的队形保持策略能实现稳定的多无人船编队导航,并在一定程度上满足队形保持的要求。相较于多智能体深度确定性策略梯度(MADDPG:Multi-Agent Depth Deterministic Policy Gradient)算法,所提出的ATMADDPG算法在收敛速度、队形保持能力和对环境变化的适应性等方面表现出更优越的性能,综合导航效率可提高约80%,具有较大的应用潜力。展开更多
In order to improve the filter accuracy for the nonlinear error model of strapdown inertial navigation system （SINS） alignment, Unscented Kalman Filter （UKF） is presented for simulation with stationary base and mo...In order to improve the filter accuracy for the nonlinear error model of strapdown inertial navigation system （SINS） alignment, Unscented Kalman Filter （UKF） is presented for simulation with stationary base and moving base of SINS alignment. Simulation results show the superior performance of this approach when compared with classical suboptimal techniques such as extended Kalman filter in cases of large initial misalignment. The UKF has good performance in case of small initial misalignment.展开更多
文摘In order to improve the filter accuracy for the nonlinear error model of strapdown inertial navigation system （SINS） alignment, Unscented Kalman Filter （UKF） is presented for simulation with stationary base and moving base of SINS alignment. Simulation results show the superior performance of this approach when compared with classical suboptimal techniques such as extended Kalman filter in cases of large initial misalignment. The UKF has good performance in case of small initial misalignment.
