A path planning approach for mobile robots using short and safe Q-learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Path planning is crucial for mobile robots, requiring navigation from start to target while avoiding obstacles.
• Traditional Q-learning methods can suffer from slow convergence and inefficient exploration.
• Ensuring both path shortness and safety is a significant challenge in mobile robotics.

Purpose of the Study:

• To develop a novel method, short and safe Q-learning (SSQL), for efficient and safe mobile robot path planning.
• To address the slow convergence issue of standard Q-learning in path planning.
• To enhance mobile robot navigation by optimizing for both path length and obstacle avoidance.

Main Methods:

• Utilized artificial potential fields to guide exploration and provide environmental priors, mitigating random exploration in Q-learning.
• Implemented a dynamic reward system to accelerate Q-learning convergence and reduce computation time.
• Evaluated the proposed SSQL method through experiments focusing on short and safe path planning scenarios.

Main Results:

• SSQL demonstrated improved performance over classical Q-learning, with marginal increases in path length (2.83-3.64%) but significant reductions in computing time (23.42-23.98%).
• The method achieved optimal path lengths in short path planning and maintained safe distances from obstacles in safe path planning.
• Comparative experiments showed SSQL outperformed other state-of-the-art algorithms in terms of path length and turning angles.

Conclusions:

• The proposed SSQL method is effective and practical for mobile robot path planning, balancing path shortness and safety.
• SSQL offers a significant improvement in convergence speed and computational efficiency compared to classical Q-learning.
• The approach provides a robust solution for complex navigation tasks, outperforming existing optimization algorithms.