Evolving self-assembly in autonomous homogeneous robots: experiments with two physical robots.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Designing controllers for autonomous robots to achieve self-assembly is challenging.
• Existing approaches often rely on pre-defined roles or explicit communication between modules.
• Mobile self-reconfigurable systems require robust coordination mechanisms.

Purpose of the Study:

• To present a homogeneous control system for self-assembling robots.
• To enable two autonomous robot modules to assemble without prior behavioral or morphological differences.
• To demonstrate role allocation based solely on robot interactions.

Main Methods:

• Utilized dynamic neural networks evolved in simulation to control robot actuators.
• Employed evolutionary robotics to design neurocontrollers.
• Tested evolved controllers on a real hardware platform (Swarm-bot).

Main Results:

• The evolved controllers successfully achieved self-assembly between two robot modules.
• The system demonstrated dynamic role specialization based on interaction alone.
• An emergent recovery mechanism improved performance, observed even without explicit reward.
• Coordination was achieved without direct communication or knowledge of other agents' states.

Conclusions:

• Self-assembling robots can coordinate effectively without explicit communication.
• Evolutionary robotics is a viable design methodology for tasks requiring fine sensory-motor coordination.
• Perceptual cues from dynamical interactions are sufficient for initiating and regulating self-assembly.