Semantic Crowd Re-purposing at TCD

TCD's crowd re-purposing system facilitates cognizant interactions of intelligent virtual characters with a semantically labelled environment. Every agent has its own configurable attributes and behaviour which alters and adapts to the context of the labelled environment. This enables the user to re-purpose the crowd to semantically similar environments. The framework supports various sources of semantically labelled data-sets - ranging from LiDAR to manually modeled virtual environments. The future work will involve generalisation of the workflow to maximise re-usability of industry-wide legacy assets.

Depicted is a crowd of autonomous agents in a virtual representation of the grounds of Trinity College, Dublin. Semantic data obtained from a LiDAR dataset of Dublin City is used to provide contextual labels to buildings and other objects in the scene, which the agents can interpret in order to guide their actions. Current work is being done to investigate how the AI behind the agents can be designed to use the semantic data to lead to convincing emergent behaviours, such as natural grouping and queueing.