To achieve behaviors that go beyond execution of preprogrammed scripts, humanoid robots are required to be implemented with cognitive abilities. Two paradigms of cognition can be implemented: deliberative and dynamic. Deliberative approaches are based on explicit representation of knowledge and are efficient in performing reasonings to achieve long term goals. On the contrary dynamical systems favor distributed knowledge and aim at robustness toward uncertainties and dynamic changes of the environment. Implementing a software solution in a humanoid robot responsible for managing the coupling of cognition and reactive responses is a complexe task. In this regard, many robot architectures, often categorized as deliberative, reactive, hybrid, behavior based or multi-agents, have been proposed. While such architectures have been successful in achieving their purpose, their adapation to other robotic systems and/or new behaviors remains difficult as requiring time and strong programming skills. Consequently, while humanoid robots are becoming accessible to a broader community, cognition implementation and behavior control remain tasks restricted to specialists.

To adress this issue, we propose Targets-Means-Drives (TDM), a behavior control architecture based on Component Based Software Engineering (CBSE) and software framework development. TDM aims at lowering the range of technical knowledge required to implement behavior control. While CBSE is widely used to achieve hardware and software integration in an elegant manner that allows code reuse, its application to a supplementary framework exclusively dedicated to behavior control and cognition implementation is novel. Because TDM can be manipulated through association of pre-programmed behavior components, it allows behavior specification through a declarative process requiring only limited programming skills.

Targets-Drives-Means Robot Architecture
http://www.ai.iit.tsukuba.ac.jp/research/tdm/