Real-time deep learning-based model predictive control of a 3-DOF biped robot leg

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Mechanical Engineering

Background:

• Traditional robotic control often relies on complex analytical dynamic models.
• Achieving precise trajectory tracking in legged robots presents significant challenges.
• Existing methods may lack the adaptability and efficiency required for dynamic robotic systems.

Purpose of the Study:

• To enhance the control of a 3 Degrees of Freedom (3-DOF) biped robot leg using deep learning.
• To develop a dynamic model for precise trajectory tracking within a Model Predictive Control (MPC) framework.
• To demonstrate the effectiveness of deep learning in improving robotic control accuracy and efficiency.

Main Methods:

• A dynamic model was created using a dataset of joint angles and actuator torques.
• The deep learning-based dynamic model was integrated into a Model Predictive Control (MPC) framework.
• Operational and safety constraints were incorporated into the MPC for robust control.

Main Results:

• The proposed deep learning model enabled precise trajectory tracking for the robot leg.
• The system achieved high accuracy and efficiency, outperforming traditional control methods.
• Experimental results validated the effectiveness of deep learning in robotic control applications.

Conclusions:

• Deep learning models can effectively enhance robotic control, particularly in achieving precise trajectory tracking.
• Integrating deep learning into MPC offers a powerful alternative to traditional dynamic models.
• This research highlights the significant potential of deep learning for advancing robotic system control and predictive techniques.