A learning and memory architecture for robot companions based on incremental associative learning

We present a learning and memory architecture that allows a robot companion to incrementally learn and associate data from different sensors and actuators. We use a topology learning algorithm that clusters the received inputs into discrete categories. On top of these clusters we apply associative learning methods to store co-occurrence relationships in an associative network. We evaluated the incremental clustering capabilities on two datasets and further performed a real experiment on associative learning where a robot learned by demonstration to associate visual perceptions with motor readings. We give a short overview of the obtained evaluation results and the experiment outcomes and we also highlight advantages of the presented architecture.