Novel deep reinforcement learning based collision avoidance approach for path planning of robots in unknown...

Area of Science:

• Artificial Intelligence
• Robotics
• Machine Learning

Background:

• Traditional motion planning algorithms struggle with real-time environmental responses, especially in cluttered spaces.
• Existing methods are computationally expensive, slow to converge, and inefficient for task learning.
• Reinforcement learning offers a promising solution through its reward-based feedback mechanisms.

Purpose of the Study:

• To develop a novel reinforcement learning algorithm for improved robot motion planning.
• To enhance the responsiveness and convergence speed of robotic navigation in complex environments.
• To integrate deep learning with Q-learning for superior path planning capabilities.

Main Methods:

• A novel Q-learning-based reinforcement learning algorithm was developed.
• Deep learning was integrated into the reinforcement learning framework.
• The algorithm was evaluated in simulated narrow and cluttered passage environments.

Main Results:

• The proposed algorithm demonstrated improved convergence rates compared to existing methods.
• In cluttered environments, the agent converged in 210 episodes.
• In narrow passages, the agent converged in 400 episodes, outperforming state-of-the-art approaches in path convergence and number of turns.

Conclusions:

• The novel Q-learning and deep learning approach significantly enhances robot motion planning efficiency.
• The algorithm provides faster convergence and better path planning in challenging, obstacle-rich environments.
• This research paves the way for more adaptive and responsive robotic navigation systems.