Design a path - planning strategy for mobile robot in multi-structured environment based on distributional...

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Mobile robots require robust navigation strategies for unknown environments.
• Image noise in depth data hinders accurate environmental perception.
• Existing path-planning methods struggle with complex, multi-object scenarios.

Purpose of the Study:

• To develop a novel, integrated path-planning framework for mobile robots in unknown environments.
• To enhance robot navigation capabilities by improving depth image quality and decision-making under uncertainty.
• To enable collision-free navigation in complex, multi-object environments.

Main Methods:

• Integration of Lightweight Learned Image Denoising with Instance Adaptation (LIDIA) for depth image denoising and calibration.
• Application of Quantile Regression Deep Q-Network (QR-DQN), a distributional reinforcement learning model, for path-planning.
• Incorporation of a sub-goal mechanism to manage complex navigation tasks.

Main Results:

• The proposed framework effectively denoises and calibrates depth images using LIDIA.
• QR-DQN successfully generates smooth, short-distance paths, outperforming other methods in various scenarios.
• The integrated approach enables effective collision-free navigation in sophisticated environmental conditions.

Conclusions:

• The combined LIDIA and QR-DQN framework offers a significant advancement in mobile robot path-planning.
• This strategy enhances robot autonomy and safety in complex, unmapped environments.
• The method demonstrates superior performance in achieving efficient and smooth path-planning.