Scene perception based visual navigation of mobile robot in indoor environment

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Indoor mobile robot navigation is crucial for applications but often costly.
• Vision-based navigation offers a cost-effective solution, leveraging deep learning for image feature abstraction.
• Existing deep scene classification networks for navigation face challenges with accuracy and computational load.

Purpose of the Study:

• To develop a low-cost, vision-only perception system for indoor mobile robot navigation.
• To improve the accuracy and efficiency of visual navigation by converting it to a scene classification problem.
• To enhance robot motion performance through an adaptive weighted control (AWC) algorithm.

Main Methods:

• Designed a shallow convolutional neural network (CNN) for efficient image processing and scene classification using a monocular camera.
• Integrated an adaptive weighted control (AWC) algorithm with regular control (RC) to optimize robot movement.
• Conducted extensive experiments in diverse, unknown indoor environments (static and dynamic).

Main Results:

• The proposed shallow CNN achieved higher scene classification accuracy and efficiency compared to existing deep networks.
• The integrated AWC and RC algorithms improved the robot's motion performance.
• The vision-based navigation system demonstrated superior capability and robustness in unknown environments.

Conclusions:

• The developed shallow CNN-based visual navigation system offers a more accurate and efficient solution for indoor mobile robots.
• The adaptive weighted control strategy enhances robot motion control, particularly in challenging environments.
• This approach provides a robust and cost-effective alternative for widespread indoor robot application.