Environmental Adaptation of Robot Morphology and Control Through Real-World Evolution

Area of Science:

• Robotics
• Artificial Intelligence
• Evolutionary Computation

Background:

• Robots require adaptability to diverse real-world environments and tasks.
• Current evolutionary robotics often relies on simplified physics simulators, limiting real-world applicability.
• Complex interactions between robot control, morphology, and environment are challenging to optimize.

Purpose of the Study:

• To evolve morphology-controller configurations for a self-reconfiguring quadruped robot using real-world evaluations.
• To assess the adaptability and generality of evolved solutions across different physical surfaces.
• To investigate the interplay between robot morphology, control, and environmental properties.

Main Methods:

• Applied evolutionary search to a mechanically self-reconfiguring quadruped robot.
• Conducted evaluations solely in real-world physical environments.
• Tested evolved solutions on initial surfaces and then on previously unseen terrains.

Main Results:

• Evolutionary search yielded high-performing and diverse morphology-controller configurations.
• Both morphology and control adapted significantly to distinct physical environments.
• Evolved parameters demonstrated transferability to novel, unseen terrains.

Conclusions:

• Real-world evolutionary evaluation is effective for developing adaptable robots.
• The method successfully optimizes both robot body and control for environmental interaction.
• The approach shows generality, enabling robots to perform in previously unencountered conditions.