基于多尺度空间表征的生物启发目标指引导航模型

Bio-inspired Goal-directed Navigation Model Based on Multi-scale Spatial Representation

为实现运行体空间认知和自主导航,借鉴生物导航机理,该文提出基于多尺度空间表征的生物启发目标指引导航模型。首先构建不同尺度位置细胞图编码空间环境,采用高斯模型模拟位置细胞放电率,并将其作为Q学习的状态输入,然后采用模拟退火方法完成行为选择,通过多次探索学习使运行体能够正确规划出一条从起始点到目标点的最短路径。仿真结果表明,该方法用于目标指引导航是可行的,相对于单尺度位置细胞空间认知模型,该方法不但符合多尺度空间表征的生物学依据,而且学习速度更快。在存在障碍物的环境中,能够顺利完成目标指引导航任务,并且当障碍物发生变化时具有较好的适应性。

In order to achieve spatial cognition and autonomous navigation, enlightened by the mechanism for biological navigation, a bio-inspired goal-directed navigation model based on a multi-scale spatial representation is proposed. First, a place cell map with different scales is constructed for encoding the space environment. Second, the firing rate of place cells in each layer is calculated by the Gaussian function as the input of Q-learning process. Third, the annealing strategy is used to choose a reasonable action. After training and learning, the robot can succeed to plan an optimal route from the starting point to the goal point. Simulation results show that, the proposed method is feasible for goal-directed navigation. Compared with the spatial cognitive model of single scale place cells, the proposed method not only meets the multi-scale spatial representation nature of place cells in hippocampus, but also has a faster learning speed. Additionally, it has good performance on completing the goaloriented navigation in the presence of obstacles, and can adapt to the change of obstacles in the environment.