Self-Organizing Map With Time-Varying Structure to Plan and Control Artificial Locomotion

Area of Science:

• Robotics
• Artificial Intelligence
• Biomimicry

Background:

• Legged robot locomotion requires sophisticated planning and control systems.
• Central pattern generators (CPGs) in biology offer a model for rhythmic movement generation.
• Learning from demonstration (LfD) is a promising approach for robot skill acquisition.

Purpose of the Study:

• To present a novel algorithm, the self-organizing map-state trajectory generator (SOM-STG), for legged robot locomotion control.
• To enable robots to autonomously generate and control diverse gaits based on learned trajectories.
• To incorporate CPG-like features for rhythmic, synchronized, and adaptable limb movements.

Main Methods:

• The SOM-STG utilizes a self-organizing map (SOM) with a time-varying structure to generate state trajectories.
• Data for training is acquired through learning by demonstration (LfD) from various agents.
• The algorithm models cyclical limb movements observed in biological locomotion, using a dog as a demonstrator agent.

Main Results:

• The SOM-STG successfully constructed multiple gaits for a simulated six-legged robot.
• The algorithm demonstrated the ability to control the robot using learned gaits and perform smooth gait transitions.
• The system learned to generate state trajectories by observing animal locomotion, showcasing biomimetic capabilities.

Conclusions:

• The SOM-STG provides an effective method for planning and controlling legged robot locomotion.
• The algorithm successfully mimics key features of biological central pattern generators.
• This approach facilitates adaptable and robust robotic movement through learning from demonstration and biomimicry.