A Bio-Inspired Mechanism for Learning Robot Motion From Mirrored Human Demonstrations

Area of Science:

• Robotics
• Neuroscience
• Machine Learning

Background:

• Human and animal studies show diverse learning mechanisms enhance skill acquisition.
• Mirror neurons activate during action execution and observation, suggesting a biological basis for motor skill reinforcement through observation.
• This implies a mechanism for modeling others' movements and integrating them with self-models for imitation.

Purpose of the Study:

• To develop a robot learning system inspired by biological mirroring mechanisms.
• To create a system capable of mapping a teaching agent's movements to a robot's configuration space.
• To refine robot skills through complementary learning examples.

Main Methods:

• A neural network integrating a motor cortex-like differential map (spiking neural networks) for task-space to joint-space transformation.
• A static map (self-organizing map) correlating the joint spaces of the robot and a teaching agent.
• Experimental validation to quantify skill improvement.

Main Results:

• The proposed neural network successfully enabled the robot to mirror human teaching agent actions in its joint-space.
• The robot's reaching skill was refined through complementary examples provided during the learning process.
• Experimental results quantified the performance improvements achieved.

Conclusions:

• The developed bio-inspired neural network facilitates robot skill acquisition through mirroring observed movements.
• This approach offers a novel method for robot learning and control, enhancing motor abilities.
• The study demonstrates the effectiveness of integrating differential and static maps for robotic imitation learning.