This paper describes a robotics cognitive architecture for social robots named CORTEX. This architecture integrates di fferent levels of abstraction (from basic geometry to high-level predicates) into a unique Deep Space Representation (DSR) that diff erent agents interface. These agents update the contents of the DSR with new data from the outer world, and execute, plan and design behaviours. The design of CORTEX as an unified deep representation allows to fit both the subsymbolic processing and exibility requirements of robot control. In this paper a first implementation of CORTEX has been integrated into Gualzru, a robotic salesman, and tested in real scenarios. Results show that this cognitive architecture allows this robot to adequately execute its use case, and that it has a promising adaptability to achieve new tasks and be used in new scenarios.
