Cognitive Architecture as a Service: Scaffolded Integration of Heterogeneous Models Through Event Streams.

The development of cognitive architectures for biomimetic robots can benefit from the seamless integration of computational models that capture some of the brain's capacity to co-ordinate adaptive behavior. Such integration could take advantage of recent advances in distributed systems technology to support the communication between models, however, a communication protocol general enough to allow for heterogeneity, yet, simple enough to be practical and widely used, remains elusive. In this work we propose a solution based on a scaffolded structure that provides constraints for the different models to satisfy. Within this paradigm, the models do not interact among themselves but communicate using event sourcing technology supported by the open source stream processing platform Apache Kafka. This design allows the integration of brain-based models without having to specify module-to-module interfaces. At the same time, the robot acts as a consumer and producer of events through the Neurorobotic Platform (NRP) (part of the Human Brain Project's EBrains platform), meaning that the cognitive architecture has the potential to integrate components provided by a growing community of computational neuroscientists, and to be integrated with different robot platforms. In this paper we present this approach, which we term Cognitive architecture as a Service (CaaS), which is further motivated by the goal of creating assistive robots for human care settings. We also describe some early results, based on the MiRo-e robot platform, aimed at the development and evaluation of brain-based control for applications in this setting.