Know Your Body Through Intrinsic Goals.

Area of Science:

• Developmental Robotics
• Computational Neuroscience
• Artificial Intelligence

Background:

• Infants explore their bodies to develop sensorimotor maps and actions crucial for cognitive and motor development.
• Current understanding often attributes early motor learning to random exploration (motor babbling).

Purpose of the Study:

• To propose and computationally model the hypothesis that early body knowledge acquisition is driven by autonomously generated goals based on intrinsic motivations.
• To investigate the role of competence-based intrinsic motivation in guiding goal formation and motor skill learning.

Main Methods:

• Developed a computational model with four components: a self-organizing neural network for sensory representations, a goal selector based on competence improvement, an echo-state network for motor control, and a goal accomplishment predictor.
• Tested the model controlling a simulated planar robot with two arms exploring its body via touch sensors.

Main Results:

• The model successfully converged to goals and developed motor skills enabling the robot to touch different parts of its own body.
• Demonstrated that body morphology influences the types of goals formed.
• Showed that competence-based intrinsic motivations significantly impact skill learning, goal selection, and even the formation of goal representations.

Conclusions:

• Autonomously generated goals, driven by intrinsic motivations and competence improvement, are a key mechanism for early body knowledge acquisition and motor development.
• The proposed computational model provides a framework for understanding how intrinsic motivation shapes learning and goal-directed behavior in embodied agents.
• Findings suggest future empirical research directions in developmental psychology and robotics.