Evolution of a predictive internal model in an embodied and situated agent

Area of Science:

• Robotics
• Artificial Intelligence
• Cognitive Science

Background:

• Robots often require continuous sensory input to perform tasks.
• Internal models are hypothesized to play a role in cognitive functions like prediction and planning.
• Understanding the emergence of internal models in artificial systems can provide insights into biological cognition.

Purpose of the Study:

• To investigate the spontaneous development of internal models in simulated robots.
• To determine if evolved robots can utilize internal models during sensory deprivation.
• To analyze the nature and function of these internal models.

Main Methods:

• Evolutionary algorithms were used to evolve simulated robots for context-dependent periodic behavior.
• Robots were tested under conditions with and without sensory stimulation.
• The internal states and sensorimotor rules of successful agents were analyzed.

Main Results:

• Evolved robots spontaneously developed internal models capable of predicting future sensory stimuli.
• These internal models allowed robots to maintain task performance during periods of sensory unavailability.
• The internal models anticipated functional properties of future states, not exact sensor readings.
• Agents exhibited consistent behavior across sensory-present and self-generated sensory phases.

Conclusions:

• Internal models can emerge for behavioral reasons and be exapted for other cognitive functions.
• Self-generated internal states may encode abstract, motor-oriented information rather than precise sensory details.
• This study provides evidence for the functional role of predictive internal models in autonomous agents.