Robots as responsible agents

The quest for real autonomous robots leads us to discuss the problem about the best possible control architecture enabling that important characteristic. It has been broadly accepted that a hybrid architecture, i.e. putting together both reactive and deliberative paradigms is needed to efficiently execute tasks in realistic dynamic environments. Our proposal, which is being implemented to control a Robuterll mobile platform, involves the use of a two-layers architecture. Using symbolic representation for knowledge and goals at the deliberative level and sub-symbolic neural networks for implementing the behaviors at the reactive level. One of the main problems we are now addressing is how to make these two levels to communicate, to interact without being completely dependent on each other. The multi-agent system framework gives a flexible strategy for single agents cooperation and enables a set of behaviours to have a certain degree of autonomy. This reactive layer works together with the cognitive control agent where goals and commitments are logically represented through simple modal logic.