论文部分内容阅读
针对连续动作和状态空间中面向目标的导航问题,依据海马结构中位置细胞相关特性和相关信息传递通路,构建海马位置细胞到前额叶皮层假设的动作细胞的脉冲神经网络模型.连续的状态空间和动作空间分别由位置细胞和动作细胞进行表征,模型采用直接强化学习与脉冲响应模型相结合的算法进行面向目标的自主导航.在Morris水迷宫环境中的仿真实验结果表明,该模型能够解决连续状态空间中面向目标导航问题,所采用算法在性能上优于传统的时间差分学习算法.调整网络中动作细胞的数量,模型的收敛性能不变,在改变状态空间和目标位置时,也可以实现面向目标的导航.
In order to solve the problem of target-oriented navigation in continuous motion and state space, a pulse neural network model of action cells in hippocampal position to the prefrontal cortical hypothesis is constructed according to the location of cells in the hippocampal formation and related information transmission paths.The continuous state space and The action space is characterized by the position cells and the action cells, respectively, and the model uses the algorithm of direct reinforcement learning and impulse response model for autonomous navigation.According to the simulation results in the Morris water maze environment, this model can solve the continuous state In the space-oriented navigation system, the proposed algorithm outperforms the traditional time-difference learning algorithm in performance.Adjusting the number of action cells in the network, the convergence performance of the model remains unchanged, and when the state space and the target location are changed, Target navigation.