Tong Guo Yes, there has been significant research into incorporating prior knowledge of object and space sizes into SLAM (Simultaneous Localization and Mapping). This approach, often referred to as "semantic SLAM" or "knowledge-informed SLAM," leverages prior information about the environment to enhance the robustness and accuracy of SLAM systems.
The idea is to input predefined information about objects or spaces, such as their dimensions or spatial relationships, into the SLAM framework to constrain or refine the map-building and localization process. For example:
Object-Aware SLAM: Research in this area focuses on recognizing objects and using their known sizes to improve map accuracy. By incorporating object models (e.g., using CAD models or shape priors), SLAM systems can better localize and map environments where these objects are present.
Semantic Mapping: This involves integrating prior knowledge of space layout, such as the dimensions of a room or expected positions of features like walls, doors, or furniture. Such information can be used as constraints during optimization to reduce errors in map construction.
Probabilistic Frameworks: Some studies embed prior knowledge within a probabilistic framework, where known object sizes or space dimensions are used as priors in Bayesian optimization, making SLAM more robust in cluttered or dynamic environments.
If you're interested, I can point you to specific papers or frameworks that explore this topic further.