In Reinforcement Learning (RL), efficiently representing the state of an observation space is a critical factor affecting an agent’s learning performance. However, most existing methods rely on discrete states or deep neural networks, which may result in insufficient expressiveness for continuous spaces and reduced computational efficiency. To address this limitation, this study introduces a state representation method based on Spatial Semantic Pointers (SSPs), which encodes observation spaces into high-dimensional vector representations. By leveraging the hexagonal grid structure inherent to SSPs, this approach provides an efficient and precise representation of continuous spaces. Compared to traditional methods, SSP-based state representation naturally handles complex continuous spaces, reduces information loss caused by discretization and demonstrates greater computational efficiency and robustness compared to neural network-based methods. In this work, we integrate our encoding method into the XLand-MiniGrid environment, which provides a challenging benchmark with a rich diversity of rulesets, further showcasing the effectiveness of our approach.