Abstract
From Flexible to habitual behaviors : neuro-inspired meta-learning for autonomous robots.
Behavioral studies of human and mammals have shown a dichotomy between two kinds of behaviours, namely habitual and goal-directed behaviors. This categorization depends mainly on the amount of training the agent receives on a certain task, and each category exhibits its particularities. In this work, we take inspiration for the computational models of habit learning behavior to apply these principles to robotics. We hypothesize that allowing a robot to learn habits on repetitive tasks would improve its behavior and ability to adapt to a real, dynamic and changing world.