Know Your Body Through Intrinsic Goals.

The first "object" that newborn children play with is their own body. This activity allows them to autonomously form a sensorimotor map of their own body and a repertoire of actions supporting future cognitive and motor development. Here we propose the theoretical hypothesis, operationalized as a computational model, that this acquisition of body knowledge is not guided by random motor-babbling, but rather by autonomously generated goals formed on the basis of intrinsic motivations. Motor exploration leads the agent to discover and form representations of the possible sensory events it can cause with its own actions. When the agent realizes the possibility of improving the competence to re-activate those representations, it is intrinsically motivated to select and pursue them as goals. The model is based on four components: (1) a self-organizing neural network, modulated by competence-based intrinsic motivations, that acquires abstract representations of experienced sensory (touch) changes; (2) a selector that selects the goal to pursue, and the motor resources to train to pursue it, on the basis of competence improvement; (3) an echo-state neural network that controls and learns, through goal-accomplishment and competence, the agent's motor skills; (4) a predictor of the accomplishment of the selected goals generating the competence-based intrinsic motivation signals. The model is tested as the controller of a simulated simple planar robot composed of a torso and two kinematic 3-DoF 2D arms. The robot explores its body covered by touch sensors by moving its arms. The results, which might be used to guide future empirical experiments, show how the system converges to goals and motor skills allowing it to touch the different parts of own body and how the morphology of the body affects the formed goals. The convergence is strongly dependent on competence-based intrinsic motivations affecting not only skill learning and the selection of formed goals, but also the formation of the goal representations themselves.

Phasic dopamine as a prediction error of intrinsic and extrinsic reinforcements driving both action acquisition and reward maximization: a simulated robotic study.

An important issue of recent neuroscientific research is to understand the functional role of the phasic release of dopamine in the striatum, and in particular its relation to reinforcement learning. The literature is split between two alternative hypotheses: one considers phasic dopamine as a reward prediction error similar to the computational TD-error, whose function is to guide an animal to maximize future rewards; the other holds that phasic dopamine is a sensory prediction error signal that lets the animal discover and acquire novel actions. In this paper we propose an original hypothesis that integrates these two contrasting positions: according to our view phasic dopamine represents a TD-like reinforcement prediction error learning signal determined by both unexpected changes in the environment (temporary, intrinsic reinforcements) and biological rewards (permanent, extrinsic reinforcements). Accordingly, dopamine plays the functional role of driving both the discovery and acquisition of novel actions and the maximization of future rewards. To validate our hypothesis we perform a series of experiments with a simulated robotic system that has to learn different skills in order to get rewards. We compare different versions of the system in which we vary the composition of the learning signal. The results show that only the system reinforced by both extrinsic and intrinsic reinforcements is able to reach high performance in sufficiently complex conditions.

Which is the best intrinsic motivation signal for learning multiple skills?

Humans and other biological agents are able to autonomously learn and cache different skills in the absence of any biological pressure or any assigned task. In this respect, Intrinsic Motivations (i.e., motivations not connected to reward-related stimuli) play a cardinal role in animal learning, and can be considered as a fundamental tool for developing more autonomous and more adaptive artificial agents. In this work, we provide an exhaustive analysis of a scarcely investigated problem: which kind of IM reinforcement signal is the most suitable for driving the acquisition of multiple skills in the shortest time? To this purpose we implemented an artificial agent with a hierarchical architecture that allows to learn and cache different skills. We tested the system in a setup with continuous states and actions, in particular, with a kinematic robotic arm that has to learn different reaching tasks. We compare the results of different versions of the system driven by several different intrinsic motivation signals. The results show (a) that intrinsic reinforcements purely based on the knowledge of the system are not appropriate to guide the acquisition of multiple skills, and (b) that the stronger the link between the IM signal and the competence of the system, the better the performance.