Autonomous Driving of Mobile Robots in Dynamic Environments Based on Deep Deterministic Policy Gradient: Reward...

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Autonomous navigation in dynamic environments presents significant challenges for mobile robots.
• Traditional methods often struggle with real-time obstacle avoidance and optimal path planning.

Purpose of the Study:

• To develop an end-to-end reinforcement learning method for autonomous mobile robot driving.
• To enhance navigation efficiency and safety in dynamic, obstacle-rich environments.

Main Methods:

• Proposed a reinforcement learning (RL)-based end-to-end learning approach.
• Integrated multifunctional reward-shaping for goal-directed guidance.
• Utilized hindsight experience replay to overcome sparse reward challenges.

Main Results:

• The proposed method demonstrated effective autonomous driving capabilities.
• Successful navigation without collisions was achieved in both simulated and real-world tests.
• Comparative experiments across five scenarios validated the technique's performance.

Conclusions:

• The combined reinforcement learning techniques enable optimal policy discovery for collision-free navigation.
• The method is robust and effective in dynamic environments with obstacles.
• This approach advances autonomous driving systems for mobile robots.