Implementation of Q learning and deep Q network for controlling a self balancing robot model

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Self-balancing robots require sophisticated control systems to maintain stability.
• Reinforcement learning offers a promising approach for developing adaptive control strategies.

Purpose of the Study:

• To implement and compare Q learning and deep Q network (DQN) algorithms for a self-balancing robot model.
• To enable the robot to learn optimal actions for environmental balance through reinforcement.

Main Methods:

• Utilized the Gazebo simulation environment to model a self-balancing robot.
• Implemented two reinforcement learning algorithms: Q learning and deep Q network (DQN).
• Conducted experiments with varying hyperparameters and analyzed performance curves.

Main Results:

• Demonstrated the feasibility of using Q learning and DQN for robot self-balancing.
• Performance curves illustrate the learning progress and effectiveness of the implemented algorithms.
• Identified optimal hyperparameters for improved balancing capabilities.

Conclusions:

• Both Q learning and DQN can effectively train a self-balancing robot.
• The choice of hyperparameters significantly impacts learning efficiency and balancing performance.
• This research provides a foundation for developing more robust autonomous balancing systems.