Multimodal Training Between Agents

In the system Locator, agents are treated as individual and autonomous subjects that are able to adapt to heterogenous user groups. Applying multimodal information from their surroundings (visual and linguistic), they acquire the necessary concepts for a successful interaction. This approach has proven successful in a domain that exhibits a remarkable variety of possible (often language-specific) structurings: the spatial domain. In this paper, the further development is described (Locator \(^{\rm {\sc 2}}\)) that allows for agent-agent interactions such that an agent is instructed by another agent that plays the role of a teacher.