仿动物大脑网格细胞神经定位机制的同步定位与地图构建方法研究

In natural world, animal’s brain has evolved considerable neural localization mechanisms to adapt various complicated environments. Researchers from biology, navigation and localization have respectively proposed a lot of neural localization models and biological inspired localization approaches, both of which can describe and mimic the neural localization mechanisms of animals to some extent. However, there are still some problems with these models, e.g., the precision of localization is low, the extensibility is weak, the navigation scale is small, the consistency with real neural mechanisms of animal is low, etc. To solve these problems, a novel Simultaneous Localization and Mapping (SLAM) approach which mimic the neural localization mechanisms of grid cells of animal’s brain is proposed in this project. The stereo-vision is provided as input to the proposed system, and a new path integration approach which can adapt to large scale environments and has better consistency with the real neural mechanisms of animal is developed through exploring and improving existing continuous attractor network models and oscillatory interference models of grid cells. A hippocampus place cell model and its relationships with grid cells and visual features are developed, and the accumulation errors associated with the proposed path integration system can be reduced by information fusion of visual features and path integration positions, so the accurate cognitive map can be built and represented. To improve the extensibility and accuracy of the model, a model parameters optimization process is proposed in the project. The project is expected to develop a novel biological inspired SLAM approach which can adapt to large scale environments, therefore a new technology for self-localization is presented; at the same time, the developed model is also expected to provide an effective validation tool for biological navigation and localization mechanisms in large scale environments.

自然界中，动物大脑经过亿万年进化形成了多种精巧的自主导航定位机制以适应各种复杂环境。生物与导航定位领域已分别提出众多神经定位模型和仿生定位方法，可在一定程度上描述和模仿大脑神经定位机制，但这些方法仍存在精度低、拓展性差、难以适应大范围环境以及与动物实际定位机理不符等问题。针对上述问题，本项目拟模仿动物大脑网格细胞神经定位机制，建立一种新的同步定位与地图构建（SLAM）方法：以立体视觉信息处理为基础，通过探索改进网格细胞连续吸引子网络模型和振荡干扰模型，建立适宜于大范围环境、符合动物神经定位机制的路径整合方法；建模海马体位置细胞及其与网格细胞、视觉特征的关联，通过视觉与路径整合信息融合消除累积误差，实现认知地图的构建与表达；开发模型参数优化方法，提高模型的精度与扩展性。项目预期实现一种新型适宜于大范围环境的仿生SLAM方法，为自主定位提供新的技术基础，同时也为生物导航定位提供大范围环境验证。

导航定位是机器人实现自主行为的关键技术，然而当前的定位方法如GPS、航迹推算等存在诸多的问题，对环境的适应性较差，极大地影响机器人自主能力的提升。在自然界中，动物大脑经过亿万年进化形成了多种精巧的自主导航定位机制以适应各种复杂环境。生物与导航定位领域已分别提出众多神经定位模型和仿生定位方法，可在一定程度上描述和模仿大脑神经定位机制，但这些方法仍存在精度低、拓展性差、难以适应大范围环境以及与动物实际定位机理不符等问题。针对上述问题，本项目提出模仿动物大脑网格细胞神经定位机制，建立一种新的同步定位与地图构建（SLAM）方法：以立体视觉信息处理为基础，通过探索改进网格细胞连续吸引子网络模型和振荡干扰模型，建立适宜于大范围环境、符合动物神经定位机制的路径整合方法；建模海马体位置细胞及其与网格细胞、视觉特征的关联，通过视觉与路径整合信息融合消除累积误差，实现认知地图的构建与表达；建立了一种结合地图匹配与仿生定位的大范围自主定位方法；提出了一种新的量子遗传算法，为仿生定位模型参数优化提供了快速有效的方法，提高模型的精度与扩展性。. 通过广泛的实验测试，验证了本项目提出的方法能够保证在无GPS条件下，自主定位精度小于10米，满足室外环境下移动机器人自主定位的需求。本项目借鉴哺乳动物的神经定位机理，研究仿生自主导航定位方法，其科学意义在于为机器人导航定位领域提供新的技术思路，从而进一步提高机器人的自主能力。

内容获取失败，请点击重试