SFPathFormer for spatial frequency robot navigation and obstacle avoidance

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Robot navigation and obstacle avoidance face challenges from perception noise, feature redundancy, and poor global modeling in complex environments.
• Existing methods struggle with safety and stability due to these limitations.

Purpose of the Study:

• To propose a vision-driven robot navigation and obstacle avoidance framework.
• To enhance path safety and planning stability in complex environments through a unified modeling pipeline.
• To address perception noise, feature redundancy, and global environment modeling deficiencies.

Main Methods:

• Spatial-frequency joint perception module using parameterized wavelet downsampling and a global spatial-frequency attention mechanism.
• Dynamic domain feature selection for path decision-making guided by contrastive learning.
• Vision Transformer for modeling global environmental structure and long-range obstacle correlations.

Main Results:

• Significantly outperforms state-of-the-art approaches in path safety, obstacle avoidance success rate, and planning stability.
• Demonstrates stronger robustness in complex backgrounds and dynamic disturbances.
• Validates synergistic effects of integrated modules for improved performance.

Conclusions:

• The proposed framework offers a scalable and effective solution for vision-driven autonomous robot navigation.
• The integration of spatial-frequency perception, dynamic feature selection, and global modeling enhances robot performance in complex environments.
• This work advances robot path planning and obstacle avoidance capabilities.