Rescue path planning for urban flood: A deep reinforcement learning-based approach

Area of Science:

• Artificial Intelligence
• Robotics
• Disaster Management

Background:

• Urban flooding is a major global disaster, necessitating efficient rescue path planning.
• Existing path planning methods struggle with risk uncertainties and complex obstacles in flood scenarios.

Purpose of the Study:

• To develop an advanced deep reinforcement learning (RL) algorithm for optimal urban flood rescue path planning.
• To address challenges of risk uncertainty and complex obstacle avoidance in dynamic flood environments.

Main Methods:

• Proposed a deep RL algorithm with dual-priority experience replays and backtrack punishment for risk estimation.
• Incorporated random noisy networks and dynamic exploration for improved obstacle navigation and area exploration.
• Utilized complex grid simulation scenarios mirroring real-world urban flood rescue operations.

Main Results:

• The proposed RL algorithm successfully bypassed all complex obstacles in nine challenging scenarios.
• Demonstrated superior performance in planning optimal rescue paths compared to existing methods.
• Validated the algorithm's adaptability to diverse and extreme obstacle configurations.

Conclusions:

• The developed RL algorithm significantly advances urban flood rescue path planning capabilities.
• Provides adaptable mechanisms for AI in navigating complex environments and enhancing real-world risk management.