E-DQN-Based Path Planning Method for Drones in Airsim Simulator under Unknown Environment.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Drone navigation in unknown environments presents significant challenges for rapid and efficient path planning.
• Traditional methods often struggle with real-time adaptation and computational efficiency.

Purpose of the Study:

• To develop a bio-inspired path planning algorithm for drones that enhances planning speed in unknown environments.
• To leverage event-based data and deep reinforcement learning for improved drone autonomy.

Main Methods:

• Utilized an event-based deep Q-network (E-DQN) by integrating event stream data into a reinforcement learning network.
• Employed an airsim simulator for collecting event data and an auto-encoder for feature extraction and event weight generation.
• Validated the algorithm in a virtual obstacle environment using Unreal Engine and airsim.

Main Results:

• The proposed E-DQN algorithm demonstrated adaptability for drones to successfully locate goals in unfamiliar terrains.
• Significant improvements in path planning rapidity were observed compared to conventional path planning methods.
• The method effectively utilizes event data for enhanced environmental perception and decision-making.

Conclusions:

• The E-DQN approach offers a promising solution for accelerating drone path planning in dynamic and unknown environments.
• This bio-inspired method enhances drone navigation capabilities, paving the way for more efficient autonomous systems.
• Event-based data processing combined with deep reinforcement learning is effective for real-time drone pathfinding.