RadDQN: A Deep Q Learning-Based Architecture for Finding Time-Efficient Minimum Radiation Exposure Pathway.

Area of Science:

• Artificial Intelligence
• Robotics
• Nuclear Engineering

Background:

• Deep reinforcement learning (DRL) is increasingly used in automation.
• Radiation-aware autonomous unmanned aerial vehicles (UAVs) face challenges with reward functions and exploration.
• Optimizing radiation exposure for UAVs in nuclear settings is critical.

Purpose of the Study:

• To develop a radiation-aware deep Q-learning network (RadDQN) for efficient, low-exposure UAV navigation.
• To address limitations in existing DRL reward functions and exploration strategies for radiation environments.

Main Methods:

• Introduced RadDQN, a novel deep Q-learning network.
• Implemented a radiation-sensitive reward function using the inverse square law.
• Developed unique exploration strategies to avoid high-radiation states.

Main Results:

• RadDQN successfully identified time-efficient, minimum radiation-exposure pathways.
• The method effectively managed diverse radiation field distributions.
• RadDQN demonstrated superior convergence rates and training stability compared to baseline models.

Conclusions:

• RadDQN offers an effective solution for radiation-aware UAV navigation in hazardous zones.
• The approach shows significant potential for real-world applications in the nuclear industry.
• This work advances DRL applications for optimizing safety and efficiency in complex environments.