Integration field-based breadth-first search for flow field pathfinding.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Science

Background:

• Flow field pathfinding is crucial for multi-agent navigation.
• Traditional methods like CFD are computationally expensive.
• Discrete flow fields often yield suboptimal, zigzag paths.

Purpose of the Study:

• To develop a more efficient and accurate flow field pathfinding method.
• To improve path length and computation speed.
• To enhance path planning adaptability in dynamic environments.

Main Methods:

• Proposed an integration field-based breadth-first search algorithm.
• Utilized wavefront parallelization for faster computation and shorter paths.
• Integrated flow fields with deep reinforcement learning for enhanced adaptability.

Main Results:

• The proposed method significantly reduces path lengths compared to traditional discrete flow fields.
• Achieved faster computation times through wavefront parallelization.
• Demonstrated superior performance of the flow field-based deep reinforcement learning framework in unknown indoor environments.

Conclusions:

• The integration field-based BFS offers an efficient and effective solution for flow field pathfinding.
• Deep reinforcement learning integration enhances navigation adaptability in dynamic settings.
• The method's practical applicability is validated in real-world environments.