Evolution of homing navigation in a real mobile robot

Area of Science:

• Robotics
• Artificial Intelligence
• Evolutionary Computation

Background:

• Mobile robots require sophisticated control systems for autonomous navigation and task completion.
• Previous methods often relied on pre-designed robot/environment interactions, limiting adaptability.
• Developing autonomous behaviors for tasks like battery charging is crucial for long-term robot operation.

Purpose of the Study:

• To describe the evolution of a discrete-time recurrent neural network (RNN) for controlling a real mobile robot.
• To demonstrate the autonomous development of behaviors for locating a battery charger and returning to it.
• To investigate the role of lifting design constraints on emergent robot behaviors.

Main Methods:

• Evolutionary procedures were conducted entirely on a physical mobile robot without human intervention.
• A discrete-time recurrent neural network was evolved to control robot behaviors.
• Robot/environment interactions were designed with fewer constraints compared to preliminary experiments.

Main Results:

• The evolved neural network enabled the mobile robot to autonomously locate its battery charger.
• The robot exhibited emergent homing behavior, periodically returning to the charger.
• An internal, non-pre-designed neural topographic map emerged, guiding trajectory selection based on location and energy levels.

Conclusions:

• Autonomous development of complex behaviors like charger homing is achievable through evolutionary procedures on physical robots.
• Lifting constraints in robot/environment design facilitates the emergence of sophisticated internal representations and control strategies.
• Evolved neural topographic maps provide an effective mechanism for energy-aware navigation in mobile robots.