An adaptive neuro-endocrine system for robotic systems

We present an adaptive artificial neural-endocrine (AANE) system that is capable of learning ldquoon-linerdquo and exploits environmental data to allow for adaptive behaviour to be demonstrated. Our AANE is capable of learning associations between sensor data and actions, and affords systems the ability to cope with sensor degradation and failure. We have tested our system in real robotic units and demonstrate adaptive behaviour over prolonged periods of time. This work is another step towards creating a robotic control system that affords ldquohomeostasisrdquo for prolonged autonomy.