Superpixel Segmentation Based Synthetic Classifications with Clear Boundary Information for a Legged Robot.

Area of Science:

• Robotics
• Computer Vision
• Machine Learning

Background:

• Autonomous multi-legged walking robots require accurate terrain classification for effective navigation.
• Existing Support Vector Machine (SVM) methods struggle with mixed terrain identification, providing only single labels.
• High-quality terrain data for legged robots is challenging to acquire.

Purpose of the Study:

• To propose novel synthetic classification methods for enhanced terrain classification in legged robots.
• To address the limitations of traditional SVM in identifying mixed terrains.
• To develop a more applicable and efficient terrain classification model for practical robotic use.

Main Methods:

• Introduction of Simple Linear Iterative Clustering based Support Vector Machine (SLIC-SVM) to handle mixed terrain classification.
• Development of Simple Linear Iterative Clustering based SegNet (SLIC-SegNet), a single-input multi-output model for improved terrain classification.
• Utilizing superpixel segmentation for robust terrain analysis.

Main Results:

• Both SLIC-SVM and SLIC-SegNet demonstrated reliable mixed terrain classification capabilities.
• The proposed methods provide clear boundary information for identified terrains.
• Experiments confirmed the effectiveness on regular, irregular, and mixed terrain types.

Conclusions:

• The developed SLIC-SegNet offers a practical solution for obtaining high-quality terrain information with reduced effort.
• These superpixel segmentation-based methods enable accurate mixed terrain classification, crucial for robotic applications.
• The findings facilitate the practical implementation of terrain-dependent gait selection and path planning for multi-legged robots.