Spatial Cognitive Maps in Agent Navigation and Path Planning

摘要

This paper presents a computational scheme for providing a human-like 2D spatial perception and memory (a computational cognitive map) for an agent and investigates the effects of memory degradations of data on navigation in closed virtual environments. An agent's (such as a mobile robot) perception of space in the form of a graph of locations with hierarchically coded distance and orientation information is investigated. The introduction of memory allows for the use of pre-stored paths for direct access instead of carrying out new searches for more informed path planning. The scheme includes associative access between locations, aggregates and uses distance and orientation information, and generally organises spatial data. It is shown that the representation carries adequate topological structure to enable the agent to find its way accurately to distant locations in its environment. The performance of the scheme is tested on a virtual agent cast in a maze-like virtual environment and the experimental results show that the process is adaptive and robust even when memory of data is degraded and remembrance of travelled locations is limited.