Computer Vision-Based Path Planning for Robot Arms in Three-Dimensional Workspaces Using Q-Learning and Neural...

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Existing path planning methods struggle with accurate 3D object localization, computational efficiency, and workspace limitations.
• Previous studies have partially addressed these issues, but significant challenges persist in real-world applications.

Purpose of the Study:

• To develop a novel path planning approach combining computer vision, Q-learning, and neural networks.
• To overcome limitations of existing methods, including unreliable 3D localization and time-consuming computations.
• To enable efficient path planning in three-dimensional workspaces.

Main Methods:

• A computer vision-neural network algorithm processed two images from different views for real-time 3D object coordinate acquisition.
• Q-learning determined optimal action sequences (up, down, left, right, backward, forward) from start to target points.
• A trained neural network translated identified actions into specific joint angles for robot control.

Main Results:

• The integrated approach achieved accurate real-time 3D object detection and localization.
• Q-learning effectively navigated the agent through the 3D workspace.
• The neural network efficiently computed joint angles, demonstrating a significant improvement over prior methods.

Conclusions:

• The novel combination of computer vision, Q-learning, and neural networks significantly alleviates limitations in 3D path planning.
• This approach offers a more robust and efficient solution for smart applications requiring precise robotic movement.
• The study demonstrates the potential of integrating deep learning and reinforcement learning for advanced robotic path planning.