sTetro-Deep Learning Powered Staircase Cleaning and Maintenance Reconfigurable Robot.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Staircase cleaning is essential for building maintenance but labor-intensive.
• Existing autonomous robots are not optimized for complex staircase environments.
• Automating staircase cleaning requires sophisticated environmental perception systems (EPS).

Purpose of the Study:

• To develop and evaluate a vision-based EPS for autonomous staircase cleaning robots.
• To enable robots to accurately perceive and navigate staircases, identifying obstacles and debris.
• To enhance the safety and efficiency of robotic building maintenance.

Main Methods:

• Proposed an operational framework for a modular robot (sTetro) using a vision-based EPS.
• Employed SSD MobileNet for real-time object detection of stairs, obstacles, and debris.
• Fused depth information with MobileNet and Support Vector Machine (SVM) to filter false staircase detections.
• Utilized contour detection and depth clustering for precise localization of steps, obstacles, and debris.
• Deployed the framework on NVIDIA Jetson Nano hardware for real-world testing.

Main Results:

• The developed framework achieved an average processing time of 310 ms.
• Demonstrated high accuracy in staircase recognition (94.32%).
• Achieved high accuracy in obstacle and debris detection (93.81%).

Conclusions:

• The proposed vision-based EPS is effective for autonomous staircase cleaning robots.
• The system enables safe navigation and efficient cleaning of multistory staircases.
• The framework offers a viable solution for automating a challenging aspect of building maintenance.