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

Flow field pathfinding is an efficient path planning method that creates a global flow field to guide agents. This approach is widely used in scenarios that require efficient and simultaneous path planning for numerous agents. Although computational fluid dynamics(CFD) can generate continuous flow fields, these methods are often computationally inefficient. Simplified methods use modified Dijkstra algorithms to compute discrete flow fields. However, these discrete fields frequently produce zigzag paths that are significantly longer than the Euclidean shortest path. To alleviate this issue, this paper proposes an integration field-based breadth-first search for flow field pathfinding. This method, based on the discrete flow field approach, uses wavefront parallelization to achieve shorter path lengths and faster computation. Furthermore, flow fields are integrated with deep reinforcement learning to enhance the adaptability of path planning in dynamic environments. Experimental results show that this flow field-based deep reinforcement learning navigation framework outperforms traditional path planning methods in indoor environments with unknown obstacles. Moreover, the feasibility of the flow field navigation method has been successfully validated in real-world environments, confirming its practical applicability.