Adaptive motivation in a biomimetic action selection mechanism

Alexandre Coninx (1, 2) , Agnès Guillot (2) , Benoît Girard (1)

1 LPPA - Laboratoire de Physiologie de la Perception et de l'Action
2 ISIR - Institut des Systèmes Intelligents et de Robotique

In this paper, we extend a basal ganglia action selection model with an adaptive motivational system. We evaluate the resulting model in a minimal survival task, where the adaptive motivational system is used to take the resource density into account. We show that the model exhibits an increased decisional autonomy, which allows the robot to behave more economically when possible, without jeopardizing its survival chances. We discuss the effect of motivational adaptation on the robot's behaviour in various initial conditions, and find that prior adaptation strongly affects selection between two unequally abundant resources. Finally, we propose hypotheses on the neurobiological mechanisms of motivational adaptation.

Mots clés: Learning and adaptation; Reward systems; decision-making; Simulation; Artificial neural networks

Domaines: Neurosciences [q-bio.NC]; Neurosciences

Alexandre Coninx, Agnès Guillot, Benoît Girard. Adaptive motivation in a biomimetic action selection mechanism. Deuxième conférence française de Neurosciences Computationnelles, "Neurocomp08", Oct 2008, Marseille, France. ⟨hal-00331576⟩