Predictive hierarchical reinforcement learning for path-efficient mapless navigation with moving target.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Deep reinforcement learning (DRL) excels at mapless robot navigation.
• Existing DRL methods struggle with moving targets, showing reduced success and efficiency.
• Mapless navigation to dynamic targets remains a significant challenge in robotics.

Purpose of the Study:

• To develop a novel framework for mapless robot navigation towards moving targets.
• To enhance navigation performance by integrating trajectory prediction into DRL.
• To address the limitations of standard DRL in dynamic environments.

Main Methods:

• Proposed the predictive hierarchical DRL (pH-DRL) framework.
• Integrated long-term trajectory prediction for enhanced planning.
• Developed the pH-DDPG algorithm using deep deterministic policy gradient for policy optimization.
• Conducted comparative experiments on the Gazebo simulator.

Main Results:

• The pH-DDPG algorithm demonstrated superior performance over other DDPG variants.
• Achieved high success rates and path efficiency in navigating to fast-moving, random targets.
• The hierarchical policy structure proved robust against prediction errors.

Conclusions:

• The pH-DRL framework offers a cost-effective and robust solution for mapless navigation to moving targets.
• Trajectory prediction is crucial for improving DRL performance in dynamic environments.
• pH-DDPG represents a significant advancement in autonomous robot navigation.