Learning to survive in a complex environment is a more relevant task for the design of intelligent autonomous machines than complex problem solving. To establish this statement, we emphasize that autonomy requires emotional-and motivational-like characteristics that are much more straightforward to define in the context of survival tasks than in problem solving which is classically considered with intelligent agents. We also propose that using a simulation platform is a good preliminary step before the design of real machines because it allows to consider another fundamental challenge, related to the association of these emotional and motivational characteristics to higher cognitive functions. These considerations are illustrated by current simulations that we are carrying out with a bio-inspired neuronal model of the cerebral architecture of primates.