Entity-Based Representation for Efficient Mapping in Robotics

By leveraging recent advancements in AI tools for object detection, we present a novel framework for robotic mapping and navigation that shifts away from traditional metric methods that require precise data. Instead, we focus on under- standing the environment through object recognition and spatial relationships rather than relying on precise positional data. We achieve this by implementing a graph of entities connected via Line of Sight, allowing the robot to navigate with minimal metric information. We tested the feasibility of this approach in a simulated environment, showing a significant reduction in the data required for navigation while successfully reaching a target object.