Learning and Acting in Peripersonal Space: Moving, Reaching, and Grasping

Area of Science:

• Robotics
• Developmental Psychology
• Artificial Intelligence

Background:

• Infants develop a sense of peripersonal space (PPS) through sensorimotor exploration.
• Understanding this process can inform autonomous robot learning.

Purpose of the Study:

• To present a computational model of infant peripersonal space (PPS) learning.
• To implement and evaluate this model on a physical robot for autonomous sensorimotor learning.

Main Methods:

• A physical robot explored its arm's configuration space and sensory inputs (proprioception, vision).
• An intrinsic motivation system drove learning by seeking reliable outcomes from unusual events.
• Knowledge was represented as a peripersonal space (PPS) graph of states and movements.

Main Results:

• The robot learned to reliably reach and move objects.
• It also made progress in learning reliable grasping, a more complex task.
• The model demonstrated autonomous learning of motion, reaching, and grasping.

Conclusions:

• The model shows promise for autonomous sensorimotor learning in robots.
• Intrinsic motivation is a key driver for learning complex interactions within peripersonal space (PPS).
• This work advances understanding of how agents learn to interact with their immediate environment.