Artificial neural network-based behavior imitation robot control device and method, and a computer-readable medium storing a computer program for providing the same

Robot control device for imitating human behavior using artificial neural networks and inverse kinematics

This technology is a behavior imitation control system that converts human joint angle data into robot joint angle data using an artificial neural network (based on fully connected layers). It uses a 6D vector, which combines the roll, pitch, and yaw data of two human joints, as input data to predict control values for robot joints with different degrees of freedom.

Due to differences in physical structure and degrees of freedom between humans and robots like humanoids, traditional geometric inverse kinematics methods require complex calculations, struggle to produce accurate results, and suffer from reduced computational efficiency because a separate neural network must be built for each individual joint.

This technology generates 6D input data by combining 6-axis angle information from specific joint pairs in human behavior data. It then outputs robot joint angle data directly through a fully connected artificial neural network configured in 6-6-12-J units, enabling behavior imitation without the need for complex inverse kinematics calculations.

‍

‍

Key Features:

• A storage unit that stores human behavior data consisting of human joint positions and the lines connecting them to form a human figure.
• An input data generation unit that creates input data by combining the relative angles of multiple human joints from the human behavior data.
• A behavior prediction unit that outputs robot joint angles by inputting data into an artificial neural network trained with human behavior data and robot behavior ground truth data constructed via inverse kinematics.
• A control unit that manages robot behavior using the outputted robot joint angle data.

‍