Deep Q-Learning in Robotics: Improvement of Accuracy and Repeatability.

Area of Science:

• Robotics and Automation
• Machine Learning Applications
• Industrial Manufacturing

Background:

• Industrial robots are crucial in manufacturing and daily life, demanding high performance in positioning accuracy, repeatability, and speed.
• Conventional methods struggle to compensate for complex robot positioning errors stemming from multiple sources.
• Machine learning (ML) offers a potential solution for improving robot positioning accuracy and expanding operational capabilities.

Purpose of the Study:

• To present an efficient, in situ methodology for real-time industrial robot position adjustment using machine learning.
• To demonstrate the application of a deep q-learning algorithm for enhancing the positioning accuracy of an articulated robot.
• To validate an ML approach that avoids large datasets and extensive computing resources.

Main Methods:

• Implementation of a deep q-learning algorithm for online robot position adjustment.
• Utilizing an articulated KUKA youBot robot for experimental validation.
• Focusing on an ML procedure that does not require extensive external datasets or high-performance computing.

Main Results:

• Significant improvement in the robot's positioning accuracy was achieved.
• Accuracy enhancement was observed after approximately 260 iterations during online operation and initial simulation.
• The deep q-learning algorithm effectively addressed positioning errors in the KUKA youBot robot.

Conclusions:

• The proposed ML methodology offers a practical and effective solution for real-time industrial robot calibration.
• Reinforced machine learning, specifically deep q-learning, can substantially improve robot positioning accuracy.
• This approach provides a valuable tool for enhancing industrial robotics performance without demanding excessive data or computational power.