Spatial and Temporal Hierarchy for Autonomous Navigation Using Active Inference in Minigrid Environment

Area of Science:

• Robotics
• Cognitive Science
• Artificial Intelligence

Background:

• Humans navigate using topological landmarks and path integration, forming hierarchical cognitive maps for efficient planning.
• Existing autonomous navigation models often lack the hierarchical structure and integrated exploration strategies seen in human behavior.

Purpose of the Study:

• To present a scalable hierarchical active inference model for autonomous navigation, exploration, and goal-oriented behavior.
• To integrate human-like navigational strategies, including curiosity-driven exploration and hierarchical environmental representation, into an AI model.

Main Methods:

• Developed a hierarchical active inference model utilizing visual and motion perception.
• Implemented a multi-level motion planning approach (context, place, motion) for navigation.
• Validated the model through simulations in a mini-grid environment.

Main Results:

• The model successfully combined curiosity-driven exploration with goal-oriented behavior.
• Demonstrated efficient navigation in novel environments and rapid progress toward targets.
• The hierarchical structure facilitated effective planning and navigation strategies.

Conclusions:

• The proposed model offers a new solution for autonomous navigation and exploration by incorporating human-like hierarchical strategies.
• This approach enhances the efficiency and adaptability of autonomous systems in complex environments.
• Further research can explore real-world applications and more complex navigation tasks.