Improved reinforcement learning path planning algorithm integrating prior knowledge.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Mobile robot navigation requires efficient path planning, especially with incomplete environmental data.
• Traditional Q-learning algorithms face challenges with slow convergence and low learning efficiency in complex environments.

Purpose of the Study:

• To optimize autonomous navigation for mobile robots with partial environmental knowledge.
• To address the slow convergence and low learning efficiency of existing path planning algorithms.

Main Methods:

• An improved Q-learning reinforcement learning algorithm incorporating prior knowledge was developed.
• Prior knowledge was used to initialize Q-values, guiding the agent towards the target.
• A dynamically adjusted greedy factor (ε) was implemented to balance exploration and exploitation.

Main Results:

• The improved Q-learning algorithm demonstrated a faster convergence rate compared to traditional methods.
• Enhanced learning efficiency was observed in mobile robot path planning simulations.
• The algorithm effectively guided the agent, reducing invalid iterations.

Conclusions:

• The proposed improved Q-learning algorithm significantly enhances the efficiency of mobile robot autonomous navigation.
• Initialization with prior knowledge and dynamic ε adjustment are key to accelerating convergence.
• This approach offers practical improvements for real-world mobile robot applications.