Is a robot that can autonomously play soccer intelligent?

摘要

Cognition theory, and cognitics, the corresponding domain for related automated techniques, provide the framework for the formal definition, the metric assessment, and the operationalization of intelligence; intelligence is the property of systems that learn, is evaluated in lin/s/bit, and is implemented in a variety of physical structures, including biological neurones, mechanical systems and computers. Autonomy implies self support and is mostly considered here for energy and decision-making domains. Games and soccer playing have been recognized as stimulating testbeds for new concepts in IAS field. We at eivd have designed various systems in the past for robotic competitions. Diego{sup}3 for instance has been competing in official, reduced-sized soccer tournaments. In the same way as any model is valid only for some bounded domain of reality, all associated cognitive properties, including the one of intelligence are similarly limited. But within this boundary, and specifically soccer-playing according to European robot cup rules, interesting conclusions can be drawn. In practice what counts ultimately is not to be intelligent, i.e. able to learn but to be expert, i.e. to do right and fast; for the case of intelligence, we found Diego{sup}3 intelligent enough in competition, but comparatively more limited for inter-match strategy adaptation. The successor, Arthur, was made more intelligent by the design of our Piaget interpreter.