Develop robotic localisation and mapping for an object-based environment
Current localisation and mapping techniques rely either on tracking low-level geometric features in the environment, such as corners and edges, or on dense mapping of the static world.
In contrast, our perception objective provides a higher-level understanding of the environment by considering the semantic properties of individual objects, their co-observability, and placement.
We will extend existing localisation and mapping techniques to also exploit the rich contextual information afforded by the object-based representation.
This will involve a significant reconsideration of our traditional feature-based formulations to fully leverage object-based associations in a mathematically sound framework.